slack-radar

\n\nA PM-shaped agent for Slack channel sprawl. Leads cannot read forty channels, so\ndecisions about their area get made in rooms they are not in. slack-radar reads\nthose rooms and tells them when something is worth their attention.\n\n## What it does\n\n**Continuously, in every watched channel:**\n\nRecords each message and folds it into a *discussion* (a thread, or a top-level\nmessage and whatever follows it). No model runs on this path — ingest is an\nINSERT, which is what makes watching dozens of channels affordable.\n\n**On each `discussion_sweep` run:**\n\n1. **Picks ripe discussions** — a thread that has been quiet for ~20 minutes, or\n has crossed 8 messages while still live. Never one that just started.\n2. **Shortlists teams** with a free lexical pass over topics and keywords. A\n team is never a candidate in one of its own home channels — it is already in\n the room.\n3. **Judges** with one model call per discussion, scoring every shortlisted team\n at once against four signal types: a *decision* forming, an *unanswered\n question*, an *incident*, or a thread *escalating*.\n4. **Notifies** — urgent items DM the lead immediately (inside working hours\n only); everything else waits for the digest. A (discussion, team) pair is\n only ever raised once.\n\n**On each `lead_digest` run:**\n\nSends each lead one grouped message with everything pending, highest confidence\nfirst, each with a link straight to the thread.\n\n**Neither job has a default schedule.** Both are entered at deploy time and the\nagent does nothing until they are. The cadences below are what this design\nassumes, not what you get:\n\n| Job | Suggested | Why |\n|---|---|---|\n| `discussion_sweep` | `*/15 * * * *` | The real latency floor — nothing is judged sooner than the next tick. `*/5` if 15 minutes is too slow for incidents, at 3x the pod churn. |\n| `lead_digest` | `0 9,14 * * 1-5` | Start of morning and after lunch. **Keep it inside working hours:** this job has no working-hours gate of its own, so an hourly cron DMs every lead at 03:00. |\n\n`ast agent redeploy --id --adapter slack --schedule =''`\nsets or changes them. `--adapter slack` is required every time: it defaults to\n`web`, and omitting it drops Slack ingestion silently.\n\n**Always:**\n\nLeads react :+1: or :-1: on any notification. A (team, channel) pairing that\nkeeps getting thumbs-downed has to clear a higher confidence bar next time, so\nthe radar gets quieter where it is wrong without anyone editing config.\n\n**On @mention or DM:**\n\n- \"who owns X?\" — resolves an area to a team and its leads\n- \"what did I miss?\" — recent flagged discussions for your teams\n- \"has anyone mentioned X?\" — literal search across everything observed\n- \"also watch me for X\" / \"stop sending me X\" — edits your subscription live\n- \"mute me for the afternoon\" — pauses delivery\n\n## What gets you pinged, and when\n\n**15 minutes is the floor for everything** — set by the sweep cadence, not by any\nthreshold. Nothing is judged instantly, by design: a thread has to settle before\nit is worth an opinion.\n\n| You post | Outcome |\n|---|---|\n| \"the api gateway is down, anyone on it?\" | `incident`, high urgency → **DM in ~15 min** (weekday 09:00-18:00; otherwise held for the digest) |\n| \"should we move rate limiting into the gateway?\" | `decision` → **DM if the choice is landing**, else digest. The case this agent exists for. |\n| \"what's the link to our API gateway?\" | trivial ask → **digest at most**. If this pings you, the bar is too low. |\n| \"anyone up for lunch?\" | no keyword overlap → **no model call, no cost, nothing** |\n| Anything in a channel your team already lives in | **nothing, ever** — it is not news to you (`home_channels`) |\n| A thread you were already pinged about, now 30 messages | **no second DM** — matches are raise-once |\n| A CI or GitHub app posting an alert | **invisible** — the platform drops bot-authored messages |\n\nA thread with replies flowing trips the burst rule and is judged on the next\nsweep regardless of how long it has been quiet, so real incidents are fast while\na lone unanswered message waits for the silence window. Full gate-by-gate rules,\ntimings and a \"why didn't I get pinged\" checklist are in `README.md`.\n\n## Containers\n\n| Container | Trigger | Role |\n|---|---|---|\n| `agent` | always-on | Records every observed message (no model call). Answers leads on @mention/DM. Runs schema DDL and bootstraps the registry from `TEAMS_CONFIG` if it is empty. |\n| `discussion_sweep` | schedule (set at deploy) | Ripeness, prefilter, judge, raise matches, urgent DMs, feedback collection, retention purge. |\n| `lead_digest` | schedule (set at deploy) | Per-lead rollup of everything pending. |\n\n## Configuration\n\nNothing operational lives in the repo — this is a blueprint. Watched channels\nare set on the deploy page (**Observe Channel IDs** in the Slack section), and\nthe team registry lives in Postgres.\n\nBootstrap it by pasting JSON (or YAML) into the `TEAMS_CONFIG` deploy input:\n\n```json\n{\"teams\":[\n {\"key\":\"platform\",\n \"name\":\"Platform\",\n \"description\":\"Owns the API gateway, auth/SSO and rate limiting.\",\n \"leads\":[\"U01ABCDEFGH\"],\n \"keywords\":[\"gateway\",\"envoy\",\"rate limit\",\"sso\",\"429\"],\n \"topics\":[\"API gateway routing and rate limits\"],\n \"home_channels\":[\"C01ABCDEFGH\"]}\n]}\n```\n\nOnly `key` is required. The full field list is `key`, `name`, `description`,\n`leads`, `topics`, `keywords`, `home_channels`, `realtime`, `min_confidence` —\n**anything else is ignored without warning**, which is the usual reason a\nhand-written config does nothing. `leads` are Slack user IDs (`U…`), not\n`@handles`, and the example placeholders are rejected on purpose.\n\nOr skip the input entirely and DM the agent: *\"set up a team for platform, I am\nthe lead, we own the API gateway\"*.\n\n`TEAMS_CONFIG` is applied only when the registry is empty, so a redeploy never\nreverts what leads have tuned over Slack.\n\nAlways fill in a team's **home channels** — channels they already sit in are\nnever flagged for them, and it is the single most effective noise control here.\nSee `README.md` for the annotated field table, or `teams.example.yml` for a\nworked example.\n\n## Environment\n\n| Variable | Source |\n|---|---|\n| `POSTGRES_HOST/PORT/USER/PASSWORD/DB` | `slackradardb` postgres knowledge provider |\n| `ANTHROPIC_API_KEY` | top-level input — used by the agent AND the judge |\n| `BASETEN_API_KEY` | top-level input — set it to move the whole agent to Baseten |\n| `BASETEN_BASE_URL` / `BASETEN_MODEL` | top-level inputs — endpoint and model for the Baseten path |\n| `RADAR_SLACK_BOT_TOKEN` | ingestion input — provider vars do not reach ingestion containers |\n| `SLACK_WORKSPACE_DOMAIN` | top-level input, for building thread deep links |\n\n## Model backends\n\nTwo paths, one switch. Leave `BASETEN_API_KEY` blank and everything runs on\nAnthropic with `ANTHROPIC_API_KEY`; set it and everything — the Slack agent and\nthe relevance judge — runs on Baseten. Never a mix: the same rule is applied in\nboth containers, and Baseten wins if both keys are present.\n\nDefaults are `claude-haiku-4-5` and `zai-org/GLM-4.7`. Both are deliberately\nsmall: the judge runs once per ripe discussion, and the prompt plus the\nprefilter do most of the precision work. Override with `ANTHROPIC_MODEL` /\n`BASETEN_MODEL`, or `JUDGE_MODEL` to change only the judge.\n\nTuning (ripeness thresholds, prefilter strictness, working window, retention) is\nhardcoded in `scheduler/src/config.ts` rather than exposed at deploy time — see\nthe table in `README.md`. Deploy asks only for credentials and workspace locale.\n\n## Slack app setup\n\nOAuth scopes: `channels:history`, `groups:history` (private channels),\n`chat:write`, `im:write`, `reactions:read`, `users:read`.\n\nTwo separate requirements per channel, and missing either is silent: invite the\nbot to the channel, **and** list the channel ID in the deploy page's **Observe\nChannel IDs**.\n\nNote the messaging sidecar drops any message carrying a Slack `bot_id`, so\nalerts posted by other apps are invisible; a discussion started by a bot alert is\nonly seen from the first human reply onward.\n\n## Data handling\n\nRaw message text is stored in Postgres to make batched judging possible, and\ndeleted after `MESSAGE_RETENTION_DAYS` (default 30). Transcripts sent to the\nmodel are anonymised to `person1`, `person2` — the judge does not need to know\nwho is speaking to decide whether a subject is a team's business. Matches and\nnotifications hold headlines, rationales and the one-line \"where it landed\"\nsummary — model-written prose about a thread, never message bodies. Those outlive the raw\ntext they were derived from, which is the point: a digest stays readable after\nthe transcript behind it has been purged.\n","https://github.com/Simwar/slack-radar","notifications",{"_1039":1040,"_465":1041,"_450":1042},{"_450":1026,"_1027":1028},"ingestion",{"_1006":1007,"_1008":1009},{"_966":967,"_968":969,"_970":971,"_972":973,"_974":975,"_976":977,"_978":979,"_980":981},"knowledge",{"_963":964},{"_642":961},{"_422":962,"_426":645},[646],"slackradardb",{"_426":965},"postgres","ANTHROPIC_API_KEY",{"_439":440,"_379":1005,"_61":966,"_984":24,"_445":24},"ANTHROPIC_MODEL",{"_439":440,"_642":646,"_379":1004,"_61":968},"BASETEN_API_KEY",{"_439":440,"_379":1003,"_61":970,"_984":24,"_445":24},"BASETEN_BASE_URL",{"_439":440,"_642":1001,"_379":1002,"_61":972,"_984":24},"BASETEN_MODEL",{"_439":440,"_642":988,"_379":989,"_442":990,"_61":974,"_984":24,"_991":992},"RADAR_TIMEZONE",{"_439":440,"_642":986,"_379":987,"_61":976},"SLACK_WORKSPACE_DOMAIN",{"_439":440,"_379":985,"_61":978},"TEAMS_CONFIG",{"_439":440,"_379":982,"_442":983,"_61":980,"_984":24},"Optional. Your teams, and who to notify about what. Leave blank to set it up by DMing the agent instead. JSON or YAML — prefer JSON, which survives being flattened to one line. Only `key` is required; unknown fields are ignored. Applied only when the registry is empty; after that, edit teams in Slack. Use real Slack IDs, placeholders are rejected. Example: {\"teams\":[{\"key\":\"platform\",\"name\":\"Platform\",\"description\":\"Owns the API gateway, auth/SSO and rate limiting.\",\"leads\":[\"U01ABCDEFGH\"],\"keywords\":[\"gateway\",\"envoy\",\"rate limit\",\"sso\",\"429\"],\"topics\":[\"API gateway routing and rate limits\"],\"home_channels\":[\"C01ABCDEFGH\"]}]}","long-text","optional","The 'acme' in acme.slack.com. Without it, thread links degrade to a slack:// URI.","America/New_York","Timezone for the instant-DM window (weekdays 09:00-18:00 local). Anything outside it waits for the next digest.","zai-org/GLM-4.7","Model for the Baseten path. Every option supports tool calling, which this agent requires. Flash/Fast tiers are cheaper.","select","options",[993,994,995,996,997,998,999,1000],"zai-org/GLM-5.3","zai-org/GLM-5.3-Flash","zai-org/GLM-5.2","deepseek-ai/DeepSeek-V4-Pro","deepseek-ai/DeepSeek-V4-Flash-0731","moonshotai/Kimi-K3","nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B","openai/gpt-oss-120b","https://inference.baseten.co/v1","Baseten inference base URL. Change it for a dedicated deployment.","Set this to switch the whole agent to Baseten. Baseten wins when both keys are set.","Model for the Anthropic path, for chat and triage. Raise to claude-sonnet-5 if the radar misses things worth knowing.","Used by both the agent and the sweep judge. Required unless BASETEN_API_KEY is set.","discussion-sweep",{"_1010":1020,"_420":1021,"_1013":1022},"lead-digest",{"_1010":1011,"_420":1012,"_1013":1014},"container",{"_465":1019},[1016],"trigger",{"_62":1015},"schedule",{"_439":440,"_379":1017,"_61":1018,"_445":24},"Slack bot token (xoxb-...), so this job can send digest DMs. Same value as the sweep's.","RADAR-SLACK-BOT-TOKEN","registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/slack-radar-ingestion-lead-digest:79ce013c",{"_465":1025},[1023],{"_62":1015},{"_439":440,"_379":1024,"_61":1018,"_445":24},"Slack bot token (xoxb-...), so this job can DM leads and read their reactions. Same value as the agent's.","registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/slack-radar-ingestion-discussion-sweep:79ce013c",{"_454":1033},"schedules",{"_1029":1030,"_1031":1032},"discussion_sweep","* * * * *","lead_digest","0 9,14 * * 1-5",{"_456":1034,"_458":1035},[458],{"_1036":1037},"observe_channel_ids",[1038],"C0BU9P6R5U5","healthcheck",{"_1043":1044,"_1045":1046,"_1047":1048,"_1049":1050},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/slack-radar:79ce013c",{"_452":242,"_454":24},"interval","10s","path","/health","retries",5,"timeout","5s","#cfdded","#c0dbfc","#2d5586","#85aee0","skillit","https://assets.astropods.ai/avatars/agents/rabbah/skillit.jpg",{"_468":870,"_470":871,"_472":872,"_474":475,"_476":873,"_478":874,"_480":875,"_482":1125,"_484":877},[1062],{"_366":616,"_367":616},[1061],{"_61":192,"_347":37},{"_369":1063,"_371":1064,"_373":1065,"_375":1066,"_377":1067},"a8cf096b",{"_416":1101,"_418":1102,"_958":1103,"_862":1104,"_422":1105,"_371":584},"---\ndescription: Generic dispatcher agent that loads admin-uploaded markdown skill files from Redis, routes user prompts to them, and runs them on per-skill cron schedules with optional Slack posting.\ntags:\n - dispatcher\n - skills\n - redis\n - cron\n - slack\n - mastra\n - anthropic\n - web-search\nauthors:\n - name: Rodric Rabbah\n account: rabbah\ncapabilities:\n - Serves a password-gated admin UI at /admin for managing markdown skill files\n - Stores skill files in Redis and refreshes an in-process cache via Redis pub/sub on admin changes\n - Exposes a run_skill tool the LLM calls to load the matching skill's instructions\n - Supports explicit /skill-name dispatch in chat messages to bypass LLM routing\n - Lets admins attach a per-skill cron schedule with an optional prompt and Slack channel\n - Posts scheduled run output to an admin-configured Slack channel via the platform Slack adapter\n - Persists each schedule's last run to Redis and surfaces it inline in the admin UI\n - Sweeps stale cron-fired conversations from Mastra memory hourly with a 24h TTL\n - Optionally exposes a web_search tool (powered by Tavily) for skills that need live web results\nrepository: github:rabbah/skillit\nintegrations:\n - Anthropic\n - Redis\n - Slack\n - Mastra\n - Tavily (optional)\n---\n\n## Overview\n\n`skillit` is a generic dispatcher agent. Operators upload markdown files via\nthe admin UI; each filename becomes the name of a \"skill\" whose content is\ntreated as the system instructions for handling matching user requests. The\nagent loads skills from Redis at boot, refreshes its in-process cache via\npub/sub on admin changes, and either routes a user message to the\nbest-matching skill via the `run_skill` tool or executes an explicitly-named\nskill when the user prefixes their message with `/`.\n\nEach skill can optionally be scheduled to run on cron, with the agent\ninvoked in-process via `agent.generate()` and the result persisted to Redis\n(and posted to a Slack channel if one is configured on the schedule).\n\n## Usage\n\n### Adding and editing skills\n\nSign into `/admin` with `ADMIN_PASSWORD`. Drag-and-drop one or more `.md`,\n`.markdown`, or `.txt` files into the upload area; the filename (minus\nextension) becomes the skill name. Existing skills can be deleted from the\nsame view. Changes propagate to the agent's in-process cache immediately\nvia Redis pub/sub — no restart required.\n\nA bundled `example` skill is seeded on first boot if Redis is empty and is\nauto-filtered from routing once any user-uploaded skill exists.\n\n### Chatting with the agent\n\nIn any messaging adapter (web playground, Slack channel where the bot is\ninstalled, etc.):\n\n- Ask a question. The LLM picks the best matching skill from the dynamic\n system prompt, calls `run_skill` to load it, then follows the loaded\n instructions.\n- Prefix a message with `/` to dispatch directly, skipping LLM\n routing. The skill content is inlined into the prompt and applied as-is.\n\n### Scheduling skills\n\nFor each skill, click **Schedule** in the admin UI and provide:\n\n- **Cron expression** — a standard 5-field expression (e.g. `0 9 * * 1-5`).\n- **Prompt** (optional) — sent as the user input to the skill on each fire.\n Leave blank to fire the skill with no additional input.\n- **Timezone** (optional) — IANA TZ name; defaults to the container's TZ.\n- **Slack channel** (optional) — a Slack channel ID (`C0123456789`) to post\n the run's output to. Requires the platform's Slack adapter to be\n configured separately; the bot must be a member of the target channel.\n\nThe most recent run for each schedule is shown inline (timestamp, duration,\noutput or error). Use the **Run now** button to fire a schedule on demand.\n\n### Web search\n\nSkills can use the `web_search` tool to fetch live results from the web via\n[Tavily](https://app.tavily.com). The tool is only available when\n`TAVILY_API_KEY` is set — if it is not configured, the tool is silently\nomitted and skills that reference it will fall back to the model's training\ndata.\n\nTo enable it, set `TAVILY_API_KEY` while deploying the agent blueprint.\n\n### Inputs\n\n| Input | Required | Description |\n|---|---|---|\n| `ADMIN_PASSWORD` | yes | Password gating the `/admin` upload UI |\n| `SESSION_SECRET` | yes | HMAC secret used to sign admin session cookies |\n| `TAVILY_API_KEY` | no | Tavily API key for web search; if omitted, `web_search` is disabled |\n\n## Limitations\n\n- **Single replica only.** The scheduler runs in-process; with multiple\n replicas the same cron job fires on each.\n- **Mastra memory is in-process.** Each cron fire creates a fresh thread\n in `LibSQLStore({ url: ':memory:' })`. An hourly sweep deletes threads\n older than 24h, but they don't survive a process restart.\n- **No history of cron runs.** Only the most recent run per schedule is\n persisted in Redis (`skillit:last-run`). Older runs are overwritten.\n- **Slack posting requires the bot to be in the target channel.** Posts\n to channels the bot isn't a member of are silently dropped with\n `not_in_channel`. Use the channel ID (e.g. `C0123456789`), not the\n channel name.\n- **Session storage is ephemeral.** Admin sessions are HMAC-signed\n cookies — they survive process restarts only if `SESSION_SECRET` is\n set to a stable value (otherwise an ephemeral secret is generated and\n sessions reset on restart).\n- **Output truncation.** Cron run output is truncated at 8 KiB for\n Redis storage and ~35 KB for Slack posts.\n",{"_379":1068,"_381":1069,"_559":1070,"_383":1071,"_385":1072,"_387":1073,"_389":1074},"2026-05-23T19:55:25Z","Generic dispatcher agent that loads admin-uploaded markdown skill files from Redis, routes user prompts to them, and runs them on per-skill cron schedules with optional Slack posting.",[1097,1098,1084,1099,458,579,568,1100],[1096],{"_62":574,"_409":1095},[1086,1087,1088,1089,1090,1091,1092,1093,1094],"\n## Overview\n\n`skillit` is a generic dispatcher agent. Operators upload markdown files via\nthe admin UI; each filename becomes the name of a \"skill\" whose content is\ntreated as the system instructions for handling matching user requests. The\nagent loads skills from Redis at boot, refreshes its in-process cache via\npub/sub on admin changes, and either routes a user message to the\nbest-matching skill via the `run_skill` tool or executes an explicitly-named\nskill when the user prefixes their message with `/`.\n\nEach skill can optionally be scheduled to run on cron, with the agent\ninvoked in-process via `agent.generate()` and the result persisted to Redis\n(and posted to a Slack channel if one is configured on the schedule).\n\n## Usage\n\n### Adding and editing skills\n\nSign into `/admin` with `ADMIN_PASSWORD`. Drag-and-drop one or more `.md`,\n`.markdown`, or `.txt` files into the upload area; the filename (minus\nextension) becomes the skill name. Existing skills can be deleted from the\nsame view. Changes propagate to the agent's in-process cache immediately\nvia Redis pub/sub — no restart required.\n\nA bundled `example` skill is seeded on first boot if Redis is empty and is\nauto-filtered from routing once any user-uploaded skill exists.\n\n### Chatting with the agent\n\nIn any messaging adapter (web playground, Slack channel where the bot is\ninstalled, etc.):\n\n- Ask a question. The LLM picks the best matching skill from the dynamic\n system prompt, calls `run_skill` to load it, then follows the loaded\n instructions.\n- Prefix a message with `/` to dispatch directly, skipping LLM\n routing. The skill content is inlined into the prompt and applied as-is.\n\n### Scheduling skills\n\nFor each skill, click **Schedule** in the admin UI and provide:\n\n- **Cron expression** — a standard 5-field expression (e.g. `0 9 * * 1-5`).\n- **Prompt** (optional) — sent as the user input to the skill on each fire.\n Leave blank to fire the skill with no additional input.\n- **Timezone** (optional) — IANA TZ name; defaults to the container's TZ.\n- **Slack channel** (optional) — a Slack channel ID (`C0123456789`) to post\n the run's output to. Requires the platform's Slack adapter to be\n configured separately; the bot must be a member of the target channel.\n\nThe most recent run for each schedule is shown inline (timestamp, duration,\noutput or error). Use the **Run now** button to fire a schedule on demand.\n\n### Web search\n\nSkills can use the `web_search` tool to fetch live results from the web via\n[Tavily](https://app.tavily.com). The tool is only available when\n`TAVILY_API_KEY` is set — if it is not configured, the tool is silently\nomitted and skills that reference it will fall back to the model's training\ndata.\n\nTo enable it, set `TAVILY_API_KEY` while deploying the agent blueprint.\n\n### Inputs\n\n| Input | Required | Description |\n|---|---|---|\n| `ADMIN_PASSWORD` | yes | Password gating the `/admin` upload UI |\n| `SESSION_SECRET` | yes | HMAC secret used to sign admin session cookies |\n| `TAVILY_API_KEY` | no | Tavily API key for web search; if omitted, `web_search` is disabled |\n\n## Limitations\n\n- **Single replica only.** The scheduler runs in-process; with multiple\n replicas the same cron job fires on each.\n- **Mastra memory is in-process.** Each cron fire creates a fresh thread\n in `LibSQLStore({ url: ':memory:' })`. An hourly sweep deletes threads\n older than 24h, but they don't survive a process restart.\n- **No history of cron runs.** Only the most recent run per schedule is\n persisted in Redis (`skillit:last-run`). Older runs are overwritten.\n- **Slack posting requires the bot to be in the target channel.** Posts\n to channels the bot isn't a member of are silently dropped with\n `not_in_channel`. Use the channel ID (e.g. `C0123456789`), not the\n channel name.\n- **Session storage is ephemeral.** Admin sessions are HMAC-signed\n cookies — they survive process restarts only if `SESSION_SECRET` is\n set to a stable value (otherwise an ephemeral secret is generated and\n sessions reset on restart).\n- **Output truncation.** Cron run output is truncated at 8 KiB for\n Redis storage and ~35 KB for Slack posts.\n",[1075,1076,1077,1078,1079],{"_59":568,"_61":569,"_397":24},{"_59":1084,"_61":1085,"_397":24},{"_59":458,"_61":1083,"_397":24},{"_59":579,"_61":1082,"_397":242},{"_59":1080,"_61":1081,"_397":242},"tavily (optional)","Tavily (optional)","Mastra","Slack","redis","Redis","Serves a password-gated admin UI at /admin for managing markdown skill files","Stores skill files in Redis and refreshes an in-process cache via Redis pub/sub on admin changes","Exposes a run_skill tool the LLM calls to load the matching skill's instructions","Supports explicit /skill-name dispatch in chat messages to bypass LLM routing","Lets admins attach a per-skill cron schedule with an optional prompt and Slack channel","Posts scheduled run output to an admin-configured Slack channel via the platform Slack adapter","Persists each schedule's last run to Redis and surfaces it inline in the admin UI","Sweeps stale cron-fired conversations from Mastra memory hourly with a 24h TTL","Optionally exposes a web_search tool (powered by Tavily) for skills that need live web results","https://github.com/rabbah/skillit",{"_61":192,"_347":37},"dispatcher","skills","cron","web-search",{"_465":1114,"_420":1115,"_450":1116},{"_450":1108},{"_1084":1107},{},{"_568":1106},{"_426":568},{"_426":1084},{"_452":1109,"_454":1110},{"_463":1113},{"_456":1111,"_458":1112},[462,458],{"_460":24,"_461":24},80,"registry.astropods.ai/e9945c0f-c341-421f-8680-473e88c1807c/skillit:a8cf096b",[1117,1118,1119],{"_452":24,"_454":24},{"_439":440,"_379":1123,"_61":1124,"_445":24},{"_439":440,"_379":1121,"_61":1122,"_445":24},{"_439":440,"_379":1120,"_61":919,"_984":24,"_445":24},"Tavily API key for web search (get one at app.tavily.com). If not set, the web_search tool is disabled.","HMAC secret used to sign admin session cookies","SESSION_SECRET","Password for the admin upload UI","ADMIN_PASSWORD","#e6999d","sales-research-agent","https://assets.astropods.ai/avatars/agents/simon/sales-research-agent.jpg",{"_468":1208,"_470":1209,"_472":1210,"_474":475,"_476":1211,"_478":1212,"_480":875,"_482":1213,"_484":1214},[1132],3,{"_366":303,"_367":493},{"_369":1133,"_371":1134,"_373":1135,"_375":1136,"_377":1137},"ce35dcd6",{"_416":1150,"_418":1151,"_955":1152,"_420":1153,"_389":1154,"_958":1155,"_422":1156,"_61":1126,"_1157":1158,"_371":584},"---\ndescription: Account research assistant searches Gong call transcripts and Salesforce data to answer questions about customer accounts\nmodel: anthropic/claude-sonnet-4-5\ninterfaces: [web]\nintegrations: [gong, salesforce]\nknowledge: vectors (Qdrant)\nauthors:\n - name: Astropods\n account: Astropods\n---\n\nResearches customer accounts by combining Gong call transcript search with Salesforce CRM data. Surfaces relevant context from past conversations and deal history to help sales, CS, and leadership answer questions about any account.\n\n## What you can ask\n\n- \"What has Acme Corp said about our pricing in recent calls?\"\n- \"What's the current ARR and renewal date for Globex?\"\n- \"Who are the main contacts at Initech and what do they care about?\"\n- \"Summarize the last 3 calls with Wayne Enterprises\"\n- \"Are there any open opportunities for Umbrella Corp?\"\n- \"What competitors came up in calls with Hooli this quarter?\"\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `gongSemanticSearch` | Semantic vector search over Gong call transcript chunks in Qdrant. Supports filtering by account name, date range, and call ID. |\n| `salesforceAccountLookup` | SOSL account search with intent-based field filtering (general info, ARR, usage, team, sales context, business info) |\n| `salesforceOpportunityLookup` | Opportunity lookup by account, ID, or name with intent filtering (general, win/loss details, ARR, competitors, deal details) |\n\n## Ingestion\n\nRuns on startup. Fetches Gong calls via the Extensive Calls API (with CRM context for account name matching), pulls transcripts, chunks them with speaker awareness (~600 tokens, 75-token overlap), embeds via OpenAI `text-embedding-3-small`, and upserts into Qdrant.\n\nUse `MAX_CALLS` to cap ingestion during development (e.g. `MAX_CALLS=50`).\n\n## Salesforce authentication\n\nUses OAuth `refresh_token` flow. Configure `SF_CLIENT_ID`, `SF_CLIENT_SECRET`, and `SF_REFRESH_TOKEN` via `ast configure`.\n",{"_379":1138,"_559":1139,"_387":1140,"_389":1141},"2026-03-23T16:29:12Z","Account research assistant searches Gong call transcripts and Salesforce data to answer questions about customer accounts",[1149],"\nResearches customer accounts by combining Gong call transcript search with Salesforce CRM data. Surfaces relevant context from past conversations and deal history to help sales, CS, and leadership answer questions about any account.\n\n## What you can ask\n\n- \"What has Acme Corp said about our pricing in recent calls?\"\n- \"What's the current ARR and renewal date for Globex?\"\n- \"Who are the main contacts at Initech and what do they care about?\"\n- \"Summarize the last 3 calls with Wayne Enterprises\"\n- \"Are there any open opportunities for Umbrella Corp?\"\n- \"What competitors came up in calls with Hooli this quarter?\"\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `gongSemanticSearch` | Semantic vector search over Gong call transcript chunks in Qdrant. Supports filtering by account name, date range, and call ID. |\n| `salesforceAccountLookup` | SOSL account search with intent-based field filtering (general info, ARR, usage, team, sales context, business info) |\n| `salesforceOpportunityLookup` | Opportunity lookup by account, ID, or name with intent filtering (general, win/loss details, ARR, competitors, deal details) |\n\n## Ingestion\n\nRuns on startup. Fetches Gong calls via the Extensive Calls API (with CRM context for account name matching), pulls transcripts, chunks them with speaker awareness (~600 tokens, 75-token overlap), embeds via OpenAI `text-embedding-3-small`, and upserts into Qdrant.\n\nUse `MAX_CALLS` to cap ingestion during development (e.g. `MAX_CALLS=50`).\n\n## Salesforce authentication\n\nUses OAuth `refresh_token` flow. Configure `SF_CLIENT_ID`, `SF_CLIENT_SECRET`, and `SF_REFRESH_TOKEN` via `ast configure`.\n",[1142,1143,1144],{"_59":1148,"_61":1148,"_397":242},{"_59":1146,"_61":1147,"_397":24},{"_59":1145,"_61":1145,"_397":242},"sf","salesforce","Salesforce","gong",{"_61":272,"_347":27},{"_465":1207},{"_450":1206},{"_1201":1202},{"_1198":1199},{"_1148":1196,"_1145":1197},{"_1192":1193},{"_568":1190,"_402":1191},"providers",{"_1148":1159,"_1145":1160},{"_1161":1174,"_1163":1175},{"_1161":1162,"_1163":1164},"scope",[389],"variables",[1165,1166,1167],{"_439":440,"_379":1172,"_61":1173,"_445":24},{"_439":440,"_379":1170,"_61":1171,"_445":24},{"_439":440,"_379":1168,"_61":1169,"_445":24},"Salesforce OAuth refresh token","REFRESH_TOKEN","Salesforce connected-app consumer secret","CLIENT_SECRET","Salesforce connected-app consumer key","CLIENT_ID",[389],[1176,1177,1178,1179],{"_439":440,"_379":1188,"_61":1189,"_445":24},{"_439":440,"_379":1186,"_61":1187,"_445":24},{"_439":440,"_642":1183,"_379":1184,"_61":1185},{"_439":440,"_642":1180,"_379":1181,"_61":1182},"https://app.gong.io","Gong web app base URL (used to build call deep links)","APP_URL","https://api.gong.io","Gong API base URL","BASE_URL","Gong API access key secret","ACCESS_KEY_SECRET","Gong API access key","ACCESS_KEY",{"_426":568},{"_426":402},"vectors",{"_1194":24,"_426":1195},"persistent","qdrant",{"_426":1148},{"_426":1145},"MAX_CALLS",{"_439":671,"_379":1200,"_61":1198,"_984":24},"Maximum number of Gong calls to ingest (omit for unlimited)","startup",{"_1010":1203,"_1013":1204},{"_465":1205},{"_62":1201},"registry.astropods.ai/simon/sales-research-agent-ingestion-startup:ce35dcd6",[462,458],"registry.astropods.ai/simon/sales-research-agent:ce35dcd6","#d58e91","#fa9ea2","#f86b71","#862d30","#220b0c","#e6999f","#e08589","rest-assured-postman-mastra","registry.astropods.ai/postman","https://assets.astropods.ai/avatars/agents/postman/rest-assured-postman-mastra.jpg?v=1785441971",{"_468":1269,"_470":1270,"_472":1271,"_474":475,"_476":1272,"_478":1273,"_480":1274,"_482":1275,"_484":1276},[1224],{"_366":303,"_367":303},[1222],{"_61":1223,"_347":53},"Zain Jandali",{"_369":1225,"_371":1226,"_373":1227,"_375":1228,"_377":1229},"8ef399dc",{"_416":1248,"_418":1249,"_420":1250,"_422":1251,"_371":424},"---\ndescription: \"Converts a zipped Java REST Assured test project into a runnable Postman Collection v2.1 and publishes it.\"\ntags: [testing, migration, postman, java]\nauthors:\n - name: Zain Jandali\ncapabilities:\n - \"Accepts a multipart upload of a zipped Maven/Gradle REST Assured project and returns a published Postman collection\"\n - \"Uses a per-test-class worker agent fan-out to translate JUnit/TestNG REST Assured tests, including assertions, data-driven/parameterized rows, dynamic data, auth, and chained flows\"\n - \"Resolves base URLs / auth / secrets from property files and helpers into Postman collection variables (never inlines secrets)\"\n - \"Publishes deterministically via a single Postman API POST; the agent never publishes\"\n - \"When the source project has WireMock/mock-stub units, provisions a real hosted Postman mock server from their stubbed responses and backfills its URL into a {{mockUrl}} collection variable — best-effort, never fails the job\"\n - \"Repoints any request (mock-derived or real-test-derived) whose endpoint was also stubbed with WireMock to the hosted mock, once it exists — never touches an endpoint that has no mock coverage\"\nintegrations:\n - Postman\n - Anthropic\n---\n\n## Overview\n\nThis agent turns a Java API-test suite (REST Assured on JUnit/TestNG) into a\nrunnable Postman Collection. It runs a Mastra **coordinator** agent that resolves shared\nconfig once and fans each enumerated test unit out to a focused **worker** agent, then\nassembles one Collection v2.1 document and publishes it to Postman with a single API call.\n\nThe design is deterministic on both ends with one agentic step in the middle: unzip and\nenumerate the tests (plain code) → the agent reads the project and writes one collection\nJSON (its only side effect) → validate and publish (plain code).\n\n## Usage\n\nAsync (a conversion runs for minutes, longer than a gateway request timeout):\n\n`POST /api/workspaces/:workspaceId/convert` (`multipart/form-data`) → `202 { jobId }`\n\n- `file` (form field, required): the zipped Maven/Gradle project.\n- `collectionName` (form field or `?collectionName=`, optional): overrides the generated name.\n- `X-Postman-Api-Key` (header, required): authenticates the caller (resolved via `GET /me`) AND is used to publish. Required on `/api/jobs/:jobId` polls too.\n- `X-Anthropic-Api-Key` (header, optional): overrides the platform-injected `ANTHROPIC_API_KEY`.\n\nThe whole `/api/*` surface is authenticated by the caller's Postman API key: the agent\nresolves it via `GET /me` and authorizes only users whose team is on the allow-list\n(`ALLOWED_TEAM_IDS` in `src/auth.ts`) — invalid key → `401`, non-allow-listed team →\n`403`. Decisions are cached briefly so polling doesn't call `/me` each time. `/health`\nand the embedded demo mock stay public.\n\nThen poll `GET /api/jobs/:jobId` until `status` is `succeeded` (with `collectionUid`,\n`postmanUrl`, `testsConverted`, `testsTotal`, `unitsEnumerated`, `unitsConverted`,\n`testMethodsEnumerated`, `requestsConverted`, and optionally `mockUrl`/`mockId`/\n`mockCreationError` when the source had mock-stub units) or `failed` (with `error`).\n\nThe platform injects `ANTHROPIC_API_KEY` (via the `anthropic` model provider). `MODEL`,\n`CONVERT_TIMEOUT_MS`, and `ENABLE_SHELL_TOOL` are optional inputs with safe defaults.\n\n## Limitations\n\n- Local runtime only: the agent converts the project inside its own container.\n- `testsTotal` is the agent's self-reported count; `testsConverted` is computed deterministically from the artifact and is the source of truth.\n- The bundled demo mock answers only specific demo paths; a customer's real target APIs are unaffected.\n- The `shell` tool is disabled by default (the uploaded project is untrusted).\n- Mock-server creation only covers endpoints the project already explicitly stubbed with WireMock — it does not synthesize responses for real REST Assured test endpoints — and only fires when the whole conversion published as a single collection (no sharding).\n",{"_379":1230,"_381":1231,"_559":1232,"_385":1233,"_387":1234,"_389":1235},"2026-09-03T20:06:06Z","Converts a zipped Java REST Assured test project into a runnable Postman Collection v2.1 and publishes it.",[1245,1246,33,1247],[1244],[1238,1239,1240,1241,1242,1243],"\n## Overview\n\nThis agent turns a Java API-test suite (REST Assured on JUnit/TestNG) into a\nrunnable Postman Collection. It runs a Mastra **coordinator** agent that resolves shared\nconfig once and fans each enumerated test unit out to a focused **worker** agent, then\nassembles one Collection v2.1 document and publishes it to Postman with a single API call.\n\nThe design is deterministic on both ends with one agentic step in the middle: unzip and\nenumerate the tests (plain code) → the agent reads the project and writes one collection\nJSON (its only side effect) → validate and publish (plain code).\n\n## Usage\n\nAsync (a conversion runs for minutes, longer than a gateway request timeout):\n\n`POST /api/workspaces/:workspaceId/convert` (`multipart/form-data`) → `202 { jobId }`\n\n- `file` (form field, required): the zipped Maven/Gradle project.\n- `collectionName` (form field or `?collectionName=`, optional): overrides the generated name.\n- `X-Postman-Api-Key` (header, required): authenticates the caller (resolved via `GET /me`) AND is used to publish. Required on `/api/jobs/:jobId` polls too.\n- `X-Anthropic-Api-Key` (header, optional): overrides the platform-injected `ANTHROPIC_API_KEY`.\n\nThe whole `/api/*` surface is authenticated by the caller's Postman API key: the agent\nresolves it via `GET /me` and authorizes only users whose team is on the allow-list\n(`ALLOWED_TEAM_IDS` in `src/auth.ts`) — invalid key → `401`, non-allow-listed team →\n`403`. Decisions are cached briefly so polling doesn't call `/me` each time. `/health`\nand the embedded demo mock stay public.\n\nThen poll `GET /api/jobs/:jobId` until `status` is `succeeded` (with `collectionUid`,\n`postmanUrl`, `testsConverted`, `testsTotal`, `unitsEnumerated`, `unitsConverted`,\n`testMethodsEnumerated`, `requestsConverted`, and optionally `mockUrl`/`mockId`/\n`mockCreationError` when the source had mock-stub units) or `failed` (with `error`).\n\nThe platform injects `ANTHROPIC_API_KEY` (via the `anthropic` model provider). `MODEL`,\n`CONVERT_TIMEOUT_MS`, and `ENABLE_SHELL_TOOL` are optional inputs with safe defaults.\n\n## Limitations\n\n- Local runtime only: the agent converts the project inside its own container.\n- `testsTotal` is the agent's self-reported count; `testsConverted` is computed deterministically from the artifact and is the source of truth.\n- The bundled demo mock answers only specific demo paths; a customer's real target APIs are unaffected.\n- The `shell` tool is disabled by default (the uploaded project is untrusted).\n- Mock-server creation only covers endpoints the project already explicitly stubbed with WireMock — it does not synthesize responses for real REST Assured test endpoints — and only fires when the whole conversion published as a single collection (no sharding).\n",[1236,1237],{"_59":33,"_61":233,"_397":242},{"_59":568,"_61":569,"_397":24},"Accepts a multipart upload of a zipped Maven/Gradle REST Assured project and returns a published Pos","Uses a per-test-class worker agent fan-out to translate JUnit/TestNG REST Assured tests, including a","Resolves base URLs / auth / secrets from property files and helpers into Postman collection variable","Publishes deterministically via a single Postman API POST; the agent never publishes","When the source project has WireMock/mock-stub units, provisions a real hosted Postman mock server f","Repoints any request (mock-derived or real-test-derived) whose endpoint was also stubbed with WireMo",{"_61":1223},"testing","migration","java",{"_465":1267,"_450":1268},{"_450":1265},{"_1253":1254,"_1255":1256,"_1257":1258},{"_568":1252},{"_426":568},"CONVERT_TIMEOUT_MS",{"_439":440,"_642":1263,"_379":1264,"_442":443,"_61":1253,"_984":24},"ENABLE_SHELL_TOOL",{"_439":440,"_642":1261,"_379":1262,"_442":443,"_61":1255,"_984":24},"MODEL",{"_439":440,"_642":1259,"_379":1260,"_442":443,"_61":1257,"_984":24},"claude-opus-4-8","Anthropic model id for the coordinator and worker agents","0","Set to 1 to grant the coordinator a shell tool. Untrusted input — leave off.","600000","Wall-clock cap per conversion before the run is aborted (ms)",{"_452":1266},{"_463":1113},"registry.astropods.ai/postman/rest-assured-postman-mastra:8ef399dc",{"_452":24},"#957531","#fadc9e","#c68600","#86692d","#221b0b","#f6f5f4","#d9c8a6","#e0c385","https://assets.astropods.ai/avatars/agents/zain/rest-assured-postman-mastra.jpg?v=1784145705",{"_468":1310,"_470":1311,"_472":1312,"_474":475,"_476":1313,"_478":1314,"_480":875,"_482":1315,"_484":1316},[1283],{"_366":303,"_367":493},[1282],{"_61":1223,"_347":53},{"_369":1284,"_371":1285,"_373":1286,"_375":1287,"_377":1288},"fb318cb7",{"_416":1297,"_418":1298,"_420":1299,"_422":1300,"_371":1301},"---\ndescription: \"Converts a zipped Java REST Assured test project into a runnable Postman Collection v2.1 and publishes it.\"\ntags: [testing, migration, postman, java]\nauthors:\n - name: Zain Jandali\ncapabilities:\n - \"Accepts a multipart upload of a zipped Maven/Gradle REST Assured project and returns a published Postman collection\"\n - \"Uses a per-test-class worker agent fan-out to translate JUnit/TestNG REST Assured tests, including assertions, data-driven/parameterized rows, dynamic data, auth, and chained flows\"\n - \"Resolves base URLs / auth / secrets from property files and helpers into Postman collection variables (never inlines secrets)\"\n - \"Publishes deterministically via a single Postman API POST; the agent never publishes\"\nintegrations:\n - Postman\n - Anthropic\n---\n\n## Overview\n\nThis agent turns a Java API-test suite (REST Assured on JUnit/TestNG) into a\nrunnable Postman Collection. It runs a Mastra **coordinator** agent that resolves shared\nconfig once and fans each enumerated test unit out to a focused **worker** agent, then\nassembles one Collection v2.1 document and publishes it to Postman with a single API call.\n\nThe design is deterministic on both ends with one agentic step in the middle: unzip and\nenumerate the tests (plain code) → the agent reads the project and writes one collection\nJSON (its only side effect) → validate and publish (plain code).\n\n## Usage\n\nAsync (a conversion runs for minutes, longer than a gateway request timeout):\n\n`POST /api/workspaces/:workspaceId/convert` (`multipart/form-data`) → `202 { jobId }`\n\n- `file` (form field, required): the zipped Maven/Gradle project.\n- `collectionName` (form field or `?collectionName=`, optional): overrides the generated name.\n- `X-Postman-Api-Key` (header, required): the Postman key used to publish (per request, never stored).\n- `X-Anthropic-Api-Key` (header, optional): overrides the platform-injected `ANTHROPIC_API_KEY`.\n\nThen poll `GET /api/jobs/:jobId` until `status` is `succeeded` (with `collectionUid`,\n`postmanUrl`, `testsConverted`, `testsTotal`, `unitsEnumerated`, `unitsConverted`,\n`testMethodsEnumerated`, `requestsConverted`) or `failed` (with `error`).\n\nThe platform injects `ANTHROPIC_API_KEY` (via the `anthropic` model provider). `MODEL`,\n`CONVERT_TIMEOUT_MS`, and `ENABLE_SHELL_TOOL` are optional inputs with safe defaults.\n\n## Limitations\n\n- Local runtime only: the agent converts the project inside its own container.\n- `testsTotal` is the agent's self-reported count; `testsConverted` is computed deterministically from the artifact and is the source of truth.\n- The bundled demo mock answers only specific demo paths; a customer's real target APIs are unaffected.\n- The `shell` tool is disabled by default (the uploaded project is untrusted).\n",{"_379":1230,"_381":1289,"_559":1290,"_385":1291,"_387":1292,"_389":1293},"2026-07-23T23:27:31Z",[1245,1246,33,1247],[1296],[1238,1239,1240,1241],"\n## Overview\n\nThis agent turns a Java API-test suite (REST Assured on JUnit/TestNG) into a\nrunnable Postman Collection. It runs a Mastra **coordinator** agent that resolves shared\nconfig once and fans each enumerated test unit out to a focused **worker** agent, then\nassembles one Collection v2.1 document and publishes it to Postman with a single API call.\n\nThe design is deterministic on both ends with one agentic step in the middle: unzip and\nenumerate the tests (plain code) → the agent reads the project and writes one collection\nJSON (its only side effect) → validate and publish (plain code).\n\n## Usage\n\nAsync (a conversion runs for minutes, longer than a gateway request timeout):\n\n`POST /api/workspaces/:workspaceId/convert` (`multipart/form-data`) → `202 { jobId }`\n\n- `file` (form field, required): the zipped Maven/Gradle project.\n- `collectionName` (form field or `?collectionName=`, optional): overrides the generated name.\n- `X-Postman-Api-Key` (header, required): the Postman key used to publish (per request, never stored).\n- `X-Anthropic-Api-Key` (header, optional): overrides the platform-injected `ANTHROPIC_API_KEY`.\n\nThen poll `GET /api/jobs/:jobId` until `status` is `succeeded` (with `collectionUid`,\n`postmanUrl`, `testsConverted`, `testsTotal`, `unitsEnumerated`, `unitsConverted`,\n`testMethodsEnumerated`, `requestsConverted`) or `failed` (with `error`).\n\nThe platform injects `ANTHROPIC_API_KEY` (via the `anthropic` model provider). `MODEL`,\n`CONVERT_TIMEOUT_MS`, and `ENABLE_SHELL_TOOL` are optional inputs with safe defaults.\n\n## Limitations\n\n- Local runtime only: the agent converts the project inside its own container.\n- `testsTotal` is the agent's self-reported count; `testsConverted` is computed deterministically from the artifact and is the source of truth.\n- The bundled demo mock answers only specific demo paths; a customer's real target APIs are unaffected.\n- The `shell` tool is disabled by default (the uploaded project is untrusted).\n",[1294,1295],{"_59":33,"_61":233,"_397":242},{"_59":568,"_61":569,"_397":24},{"_61":1223},{"_465":1308,"_450":1309},{"_450":1306},{"_1253":1303,"_1255":1304,"_1257":1305},{"_568":1302},"astro/v1",{"_426":568},{"_439":440,"_642":1263,"_379":1264,"_442":443,"_61":1253,"_984":24},{"_439":440,"_642":1261,"_379":1262,"_442":443,"_61":1255,"_984":24},{"_439":440,"_642":1259,"_379":1260,"_442":443,"_61":1257,"_984":24},{"_452":1307},{"_463":1113},"registry.astropods.ai/4f318af3-2951-4bda-97f1-ee29e6a4be0b/rest-assured-postman-mastra:fb318cb7",{"_452":24},"#c29386","#edb9ab","#df8269","#86402d","#22100b","#e6a999","#e09885","release-note-helper","https://assets.astropods.ai/avatars/agents/simon/release-note-helper.jpg",{"_468":1366,"_470":1367,"_472":1368,"_474":475,"_476":1369,"_478":1370,"_480":539,"_482":1371,"_484":1372},[1325],{"_366":1324,"_367":616},[1323],{"_61":31,"_347":31},14,{"_369":1326,"_371":1327,"_373":1328,"_375":1329,"_377":1330},"ef3de572",{"_416":1340,"_418":1341,"_389":1342,"_958":1343,"_422":1344,"_1157":1345,"_371":424},"---\ndescription: An agent that helps you craft release notes from Jira issues and GitHub PRs\nmodel: anthropic/claude-sonnet-4-5\ninterfaces: [web, slack]\nintegrations: [jira, github]\nknowledge: preferences (Redis)\nrepository:\n type: git\n url: \"https://github.com/astropods/agents\"\n directory: release-note-helper\n---\n\nGuides you through the full release note workflow: queries Jira for completed issues, helps you curate which ones belong in the release, verifies linked GitHub PRs, and drafts a formatted release note matching your team's style.\n\nRemembers your preferences (default project, GitHub repo, selection criteria, release note format) across sessions.\n\n## What you can ask\n\n- \"Draft release notes for PROJ issues closed in the last 2 weeks\"\n- \"Show me all Done issues for the API project since Monday\"\n- \"Skip internal issues and infrastructure changes\"\n- \"Check if the linked PRs are merged\"\n- \"Format this like our last release note\"\n- \"Save this format as my default\"\n\n## Workflow\n\n1. **Query Jira** — fetch issues moved to \"Done\" for a project and date range\n2. **Candidate selection** — agent evaluates each issue against your saved criteria\n3. **User review** — accept, reject, or override recommendations interactively\n4. **PR verification** — confirm GitHub PRs are merged and tagged\n5. **Draft** — generate a formatted release note matching your saved example\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `searchJiraIssues` | Search Jira for issues by project, status, and date range |\n| `getJiraIssueDetails` | Fetch full details for a specific issue |\n| `checkGithubPRs` | Look up linked PRs, merge status, and release tags |\n| `loadPreferences` | Read saved preferences from Redis |\n| `savePreferences` | Persist preferences (project, repo, criteria, format) to Redis |\n\n## Persistent storage\n\n- **Conversation memory** — LibSQL for chat history across sessions\n- **User preferences** — Redis for structured settings (survives restarts)\n\n## Guardrails\n\nInput prompt-injection detection via Lakera Guard.\n",{"_379":1331,"_383":1332,"_387":1333,"_389":1334},"2026-05-06T19:31:16Z","An agent that helps you craft release notes from Jira issues and GitHub PRs",{"_62":574,"_409":410,"_411":1317},"\nGuides you through the full release note workflow: queries Jira for completed issues, helps you curate which ones belong in the release, verifies linked GitHub PRs, and drafts a formatted release note matching your team's style.\n\nRemembers your preferences (default project, GitHub repo, selection criteria, release note format) across sessions.\n\n## What you can ask\n\n- \"Draft release notes for PROJ issues closed in the last 2 weeks\"\n- \"Show me all Done issues for the API project since Monday\"\n- \"Skip internal issues and infrastructure changes\"\n- \"Check if the linked PRs are merged\"\n- \"Format this like our last release note\"\n- \"Save this format as my default\"\n\n## Workflow\n\n1. **Query Jira** — fetch issues moved to \"Done\" for a project and date range\n2. **Candidate selection** — agent evaluates each issue against your saved criteria\n3. **User review** — accept, reject, or override recommendations interactively\n4. **PR verification** — confirm GitHub PRs are merged and tagged\n5. **Draft** — generate a formatted release note matching your saved example\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `searchJiraIssues` | Search Jira for issues by project, status, and date range |\n| `getJiraIssueDetails` | Fetch full details for a specific issue |\n| `checkGithubPRs` | Look up linked PRs, merge status, and release tags |\n| `loadPreferences` | Read saved preferences from Redis |\n| `savePreferences` | Persist preferences (project, repo, criteria, format) to Redis |\n\n## Persistent storage\n\n- **Conversation memory** — LibSQL for chat history across sessions\n- **User preferences** — Redis for structured settings (survives restarts)\n\n## Guardrails\n\nInput prompt-injection detection via Lakera Guard.\n",[1335,1336],{"_59":1338,"_61":1339,"_397":24},{"_59":408,"_61":1337,"_397":24},"GitHub","jira","Jira",{"_465":1365},{"_450":1362},{"_408":1360,"_1338":1361},{"_1358":1359},{"_568":1357},{"_1338":1346},{"_1161":1347,"_1163":1348},[389],[1349,1350,1351],{"_439":440,"_379":1355,"_61":1356,"_445":24},{"_439":440,"_379":1354,"_61":1185,"_445":24},{"_439":440,"_379":1352,"_61":1353,"_445":24},"Atlassian account email for Jira API authentication","EMAIL","Jira instance base URL (e.g. https://your-org.atlassian.net)","Jira API token","API_KEY",{"_426":568},"preferences",{"_426":1084},{"_426":408},{"_426":1338},{"_454":1363},{"_456":1364},[462],"registry.astropods.ai/simon/release-note-helper:ef3de572","#976fa2","#d9b6e2","#a656bb","#6f337f","#1d0d21","#ba99e6","#c692d3","recruiter-agent","registry.astropods.ai/nitinlodha","https://assets.astropods.ai/avatars/agents/nitinlodha/recruiter-agent.jpg",{"_468":870,"_470":871,"_472":872,"_474":475,"_476":873,"_478":874,"_480":875,"_482":1125,"_484":877},[1382],{"_366":1381,"_367":549},[1380],{"_61":121,"_347":51},26,{"_369":1383,"_371":1384,"_373":1385,"_375":1386,"_377":1387},"598c7e65",{"_416":1416,"_418":1417,"_958":1418,"_371":424},"---\nname: Recruiter Agent\ndescription: An intelligent recruiting copilot. Sign in, connect your Greenhouse (and optionally PDL / Gem), and the agent ranks your pipeline against any open req with multi-feature scoring, second-pass critic review, and network-aware reasoning. Comments on each candidate card feed back into the next ranking; rejection reasons from Greenhouse tune the rubric automatically.\ntags:\n - recruiting\n - hr\n - ranking\n - postman\nintegrations:\n - Greenhouse Harvest (user-supplied API key)\n - People Data Labs (optional, user-supplied)\n - Gem (optional, user-supplied)\n - GNews (optional, user-supplied)\n - Google Gmail + Calendar (optional, user-supplied OAuth)\n - Anthropic Claude (user-supplied, BYOK)\n - Slack (optional, deployer-supplied app tokens)\nvisibility: public\n---\n\n# Recruiter Agent\n\n## What it does\n\nGiven an open requisition, the agent pulls candidates from every source the user has connected, scores them on a thirteen-feature rubric, runs AI reasoning per candidate, applies a second-pass critic review, and returns the top thirty with concrete reasoning and \"network context\" observations the way a senior recruiter would frame them.\n\nIt works several ways from the web UI. **Applicant ranking** scores the people who applied to a req. **Whole-database sourcing** (\"source the best candidates across all of greenhouse for a CTO\") ranks the entire candidate database, applicants or not: every candidate is embedded into a persistent, incrementally-refreshed Qdrant index (built in the background, pre-warmed on session start), the whole index is ranked by similarity to the role, and the top matches get the full Claude deep-scoring pass. The rubric weights adapt to each specific role, derived per-JD on top of the learned per-archetype baseline, so a CTO and a CFO are scored differently.\n\nTwo more web modes invert the flow. **Match a profile** takes a resume (pasted or uploaded) and scores it against every open req as a fit %, blending semantic role match, skills overlap, and seniority, then a Claude re-rank sharpens the top openings and suggests the best-fit role or a new one worth opening. **Ideal profiles** takes 1-3 ideal profiles for a role (a resume, or a structured brief of title/skills/education/experience/justification) and returns the closest candidates in the database for each, ranked by similarity (top 30 for one ideal, 15 each for two, 10 each for three). Every result card across all modes is clickable into a LinkedIn-style profile panel (current title, experience, education, certificates, skills, LinkedIn link) with the reasoning or similarity shown alongside, and each candidate carries a US work-eligibility signal (green/amber/red) with optional location and eligibility filtering.\n\nThe recruiter can also ask for open requisitions as a table: \"give me all open reqs by location\" or \"show open roles in the USA as a table\" returns the open reqs inline with requisition id, title, location, department, compensation, and a JD summary, filtered or grouped by location, plus a CSV export link.\n\nThe agent learns continuously from three signals:\n\n- Thumbs feedback per card (numeric rubric weight updates), from the web UI or the Slack `feedback` command\n- Free-text comments per card (injected verbatim into the next ranking's reasoning prompt)\n- Greenhouse rejection reasons (mapped to feature impact via `intelligence/rejection_signal.py`)\n\n## Deploy your own copy\n\nEvery deployment is fully isolated: its own managed Postgres, its own auto-generated encryption secrets, its own users. You do not share data with any other deployment of this blueprint. Nothing needs to be pasted at deploy time; the agent generates and persists its platform secrets on first boot.\n\n```bash\n# 1. Install the CLI and sign in\ncurl -fsSL https://astropods.com/install | sh\nast login\n\n# 2. Deploy straight from the public catalog\nast deploy recruiter-agent --adapter slack --adapter web --wait\n```\n\nFor Slack, create your own Slack app in your workspace first: at https://api.slack.com/apps choose Create New App, then \"From an app manifest\", and paste the manifest below. Then install it to the workspace and paste the two tokens (`xapp-...` app token, `xoxb-...` bot token) when the deploy form asks for them.\n\nSlack app manifest (copy this; it is the same file as `doc/slack-app-manifest.json` in the source repo, reproduced here so you do not need repo access):\n\n```json\n{\n \"display_information\": {\n \"name\": \"Recruiter Agent\",\n \"description\": \"Candidate ranking copilot: shortlist any Greenhouse req, give feedback, draft outreach, all from Slack.\",\n \"background_color\": \"#10233F\"\n },\n \"features\": {\n \"app_home\": { \"messages_tab_enabled\": true, \"messages_tab_read_only_enabled\": false },\n \"bot_user\": { \"display_name\": \"recruiter-agent\", \"always_online\": true }\n },\n \"oauth_config\": {\n \"scopes\": {\n \"bot\": [\n \"app_mentions:read\", \"chat:write\", \"commands\",\n \"im:history\", \"im:read\", \"im:write\",\n \"users:read\", \"users:read.email\"\n ]\n }\n },\n \"settings\": {\n \"event_subscriptions\": { \"bot_events\": [\"app_mention\", \"message.im\"] },\n \"interactivity\": { \"is_enabled\": false },\n \"org_deploy_enabled\": false,\n \"socket_mode_enabled\": true,\n \"token_rotation_enabled\": false\n }\n}\n```\n\nOnce installed, the running agent also serves this manifest at `GET /slack-app-manifest.json` on its own URL, so you can download it directly from your deployment.\n\n### Heads up: your web URL may not provision\n\nThe platform's shared load balancer has a limited number of URL slots. Your deployment can come up healthy but never receive a public web URL (the link returns NXDOMAIN). This is a platform capacity issue, not a broken deploy. The Slack path does not use a URL slot and works on every deployment, so treat Slack as the primary surface and the web UI as a bonus if your URL provisions.\n\n### First-run data\n\nPick one of three on the deploy form:\n\n1. Nothing (default): users paste their own Greenhouse Harvest key (`setkey greenhouse ` in a Slack DM, or the web Settings page) and rank their real pipeline.\n2. `GREENHOUSE_MOCK_URL`: point Greenhouse calls at a Postman mock for demos.\n3. `USE_SAMPLE_DATA=true`: zero-Greenhouse demo mode; ranking runs against 15 bundled sample candidates.\n\nIn every case each user sets their own Anthropic key: `setkey anthropic sk-ant-...` in a Slack DM, or the web Settings page.\n\n## Using it from Slack (no web UI needed)\n\nDM the bot after install. Your account is created automatically from your Slack email.\n\n| Command | What it does |\n|---|---|\n| `help` | Command list |\n| `whoami` | Your account + which keys are set |\n| `setkey ` | DM only. Store your BYOK key: anthropic, greenhouse, pdl, gem, gnews |\n| `list jobs` | Open requisitions |\n| `rank ` | Ranked shortlist with reasoning, streamed progress |\n| `similar [count]` | Candidates from your pool with a similar background (semantic search) |\n| `feedback [comment]` | Thumbs on your latest shortlist; trains your rubric |\n| `draft [hook]` | Claude-drafted outreach email for a shortlisted candidate |\n\n## Architecture\n\n- FastAPI agent on Astro AI: web UI on port 80, plus the messaging sidecar for Slack\n- Managed Postgres for canonical state, auto-provisioned per deployment\n- Managed Qdrant + gateway embeddings for semantic candidate search (\"more like X\"), embeddings-only gateway usage so chat costs stay BYOK\n- In-process NetworkX context graph (swappable to a managed graph store post-launch)\n- Anthropic Claude (Sonnet for reasoning, Haiku for cheaper passes)\n- Postman API Catalog publish + Postman mocks/monitors for the integration plane\n\n## User model\n\nSign in with email and password on the web, or just DM the bot on Slack; identity unifies by email across both surfaces. Per-user credentials are Fernet-encrypted at rest. Every user brings their own Anthropic and Greenhouse keys, so costs land on the right account.\n",{"_379":1388,"_381":1389,"_387":1390,"_389":1391},"2026-08-12T20:19:20Z","An intelligent recruiting copilot. Sign in, connect your Greenhouse (and optionally PDL / Gem), and the agent ranks your pipeline against any open req with multi-feature scoring, second-pass critic re",[1413,1414,1415,33],"\n# Recruiter Agent\n\n## What it does\n\nGiven an open requisition, the agent pulls candidates from every source the user has connected, scores them on a thirteen-feature rubric, runs AI reasoning per candidate, applies a second-pass critic review, and returns the top thirty with concrete reasoning and \"network context\" observations the way a senior recruiter would frame them.\n\nIt works several ways from the web UI. **Applicant ranking** scores the people who applied to a req. **Whole-database sourcing** (\"source the best candidates across all of greenhouse for a CTO\") ranks the entire candidate database, applicants or not: every candidate is embedded into a persistent, incrementally-refreshed Qdrant index (built in the background, pre-warmed on session start), the whole index is ranked by similarity to the role, and the top matches get the full Claude deep-scoring pass. The rubric weights adapt to each specific role, derived per-JD on top of the learned per-archetype baseline, so a CTO and a CFO are scored differently.\n\nTwo more web modes invert the flow. **Match a profile** takes a resume (pasted or uploaded) and scores it against every open req as a fit %, blending semantic role match, skills overlap, and seniority, then a Claude re-rank sharpens the top openings and suggests the best-fit role or a new one worth opening. **Ideal profiles** takes 1-3 ideal profiles for a role (a resume, or a structured brief of title/skills/education/experience/justification) and returns the closest candidates in the database for each, ranked by similarity (top 30 for one ideal, 15 each for two, 10 each for three). Every result card across all modes is clickable into a LinkedIn-style profile panel (current title, experience, education, certificates, skills, LinkedIn link) with the reasoning or similarity shown alongside, and each candidate carries a US work-eligibility signal (green/amber/red) with optional location and eligibility filtering.\n\nThe recruiter can also ask for open requisitions as a table: \"give me all open reqs by location\" or \"show open roles in the USA as a table\" returns the open reqs inline with requisition id, title, location, department, compensation, and a JD summary, filtered or grouped by location, plus a CSV export link.\n\nThe agent learns continuously from three signals:\n\n- Thumbs feedback per card (numeric rubric weight updates), from the web UI or the Slack `feedback` command\n- Free-text comments per card (injected verbatim into the next ranking's reasoning prompt)\n- Greenhouse rejection reasons (mapped to feature impact via `intelligence/rejection_signal.py`)\n\n## Deploy your own copy\n\nEvery deployment is fully isolated: its own managed Postgres, its own auto-generated encryption secrets, its own users. You do not share data with any other deployment of this blueprint. Nothing needs to be pasted at deploy time; the agent generates and persists its platform secrets on first boot.\n\n```bash\n# 1. Install the CLI and sign in\ncurl -fsSL https://astropods.com/install | sh\nast login\n\n# 2. Deploy straight from the public catalog\nast deploy recruiter-agent --adapter slack --adapter web --wait\n```\n\nFor Slack, create your own Slack app in your workspace first: at https://api.slack.com/apps choose Create New App, then \"From an app manifest\", and paste the manifest below. Then install it to the workspace and paste the two tokens (`xapp-...` app token, `xoxb-...` bot token) when the deploy form asks for them.\n\nSlack app manifest (copy this; it is the same file as `doc/slack-app-manifest.json` in the source repo, reproduced here so you do not need repo access):\n\n```json\n{\n \"display_information\": {\n \"name\": \"Recruiter Agent\",\n \"description\": \"Candidate ranking copilot: shortlist any Greenhouse req, give feedback, draft outreach, all from Slack.\",\n \"background_color\": \"#10233F\"\n },\n \"features\": {\n \"app_home\": { \"messages_tab_enabled\": true, \"messages_tab_read_only_enabled\": false },\n \"bot_user\": { \"display_name\": \"recruiter-agent\", \"always_online\": true }\n },\n \"oauth_config\": {\n \"scopes\": {\n \"bot\": [\n \"app_mentions:read\", \"chat:write\", \"commands\",\n \"im:history\", \"im:read\", \"im:write\",\n \"users:read\", \"users:read.email\"\n ]\n }\n },\n \"settings\": {\n \"event_subscriptions\": { \"bot_events\": [\"app_mention\", \"message.im\"] },\n \"interactivity\": { \"is_enabled\": false },\n \"org_deploy_enabled\": false,\n \"socket_mode_enabled\": true,\n \"token_rotation_enabled\": false\n }\n}\n```\n\nOnce installed, the running agent also serves this manifest at `GET /slack-app-manifest.json` on its own URL, so you can download it directly from your deployment.\n\n### Heads up: your web URL may not provision\n\nThe platform's shared load balancer has a limited number of URL slots. Your deployment can come up healthy but never receive a public web URL (the link returns NXDOMAIN). This is a platform capacity issue, not a broken deploy. The Slack path does not use a URL slot and works on every deployment, so treat Slack as the primary surface and the web UI as a bonus if your URL provisions.\n\n### First-run data\n\nPick one of three on the deploy form:\n\n1. Nothing (default): users paste their own Greenhouse Harvest key (`setkey greenhouse ` in a Slack DM, or the web Settings page) and rank their real pipeline.\n2. `GREENHOUSE_MOCK_URL`: point Greenhouse calls at a Postman mock for demos.\n3. `USE_SAMPLE_DATA=true`: zero-Greenhouse demo mode; ranking runs against 15 bundled sample candidates.\n\nIn every case each user sets their own Anthropic key: `setkey anthropic sk-ant-...` in a Slack DM, or the web Settings page.\n\n## Using it from Slack (no web UI needed)\n\nDM the bot after install. Your account is created automatically from your Slack email.\n\n| Command | What it does |\n|---|---|\n| `help` | Command list |\n| `whoami` | Your account + which keys are set |\n| `setkey ` | DM only. Store your BYOK key: anthropic, greenhouse, pdl, gem, gnews |\n| `list jobs` | Open requisitions |\n| `rank ` | Ranked shortlist with reasoning, streamed progress |\n| `similar [count]` | Candidates from your pool with a similar background (semantic search) |\n| `feedback [comment]` | Thumbs on your latest shortlist; trains your rubric |\n| `draft [hook]` | Claude-drafted outreach email for a shortlisted candidate |\n\n## Architecture\n\n- FastAPI agent on Astro AI: web UI on port 80, plus the messaging sidecar for Slack\n- Managed Postgres for canonical state, auto-provisioned per deployment\n- Managed Qdrant + gateway embeddings for semantic candidate search (\"more like X\"), embeddings-only gateway usage so chat costs stay BYOK\n- In-process NetworkX context graph (swappable to a managed graph store post-launch)\n- Anthropic Claude (Sonnet for reasoning, Haiku for cheaper passes)\n- Postman API Catalog publish + Postman mocks/monitors for the integration plane\n\n## User model\n\nSign in with email and password on the web, or just DM the bot on Slack; identity unifies by email across both surfaces. Per-user credentials are Fernet-encrypted at rest. Every user brings their own Anthropic and Greenhouse keys, so costs land on the right account.\n",[1392,1393,1394,1395,1396,1397,1398],{"_59":1411,"_61":1412,"_397":242},{"_59":1409,"_61":1410,"_397":242},{"_59":1407,"_61":1408,"_397":242},{"_59":1405,"_61":1406,"_397":242},{"_59":1403,"_61":1404,"_397":242},{"_59":1401,"_61":1402,"_397":242},{"_59":1399,"_61":1400,"_397":242},"slack (optional, deployer-supplied app tokens)","Slack (optional, deployer-supplied app tokens)","anthropic claude (user-supplied, byok)","Anthropic Claude (user-supplied, BYOK)","google gmail + calendar (optional, user-supplied oauth)","Google Gmail + Calendar (optional, user-supplied OAuth)","gnews (optional, user-supplied)","GNews (optional, user-supplied)","gem (optional, user-supplied)","Gem (optional, user-supplied)","people data labs (optional, user-supplied)","People Data Labs (optional, user-supplied)","greenhouse harvest (user-supplied api key)","Greenhouse Harvest (user-supplied API key)","recruiting","hr","ranking",{"_1427":24,"_1039":1428,"_465":1429,"_420":1430,"_450":1431},{"_450":1422},{"_1419":1420,"_1192":1421},"db",{"_426":965},{"_426":1195},{"_452":1423,"_454":1424},{"_463":1426},{"_456":1425},[462],8000,"astro_ai_gateway",{"_1043":1454,"_1045":1455,"_1047":1456,"_1049":1044},"registry.astropods.ai/nitinlodha/recruiter-agent:598c7e65",[1432,1433,1434,1435,1436,1437,1438,1439],{"_452":24,"_454":24},{"_439":440,"_642":108,"_379":1452,"_61":1453,"_984":24},{"_439":440,"_642":108,"_379":1451,"_61":1122,"_984":24},{"_439":440,"_642":108,"_379":1449,"_61":1450,"_984":24},{"_439":440,"_642":108,"_379":1448,"_61":966,"_984":24},{"_439":440,"_642":108,"_379":1446,"_61":1447,"_984":24},{"_439":440,"_642":108,"_379":1444,"_61":1445,"_984":24},{"_439":440,"_642":108,"_379":1442,"_61":1443,"_984":24},{"_439":440,"_642":108,"_379":1440,"_61":1441,"_984":24},"(optional) Shared Slack workspace invite URL (https://join.slack.com/t/...). When set, the web UI shows a \"Also available in Slack\" banner so users can jump to Slack and use the bot from there. Identity unifies by email — same user_id across both surfaces. Leave blank to hide the banner.","SLACK_JOIN_URL","(optional) Public https URL of this deployment, e.g. https://recruiter-agent-chat-xxxx.agents.astropods.ai. Used for PDF download links posted to Slack.","APP_BASE_URL","(optional) Set to \"true\" to rank against bundled sample candidates with no Greenhouse account needed. Users still bring their own Anthropic key. Leave empty for real or mocked Greenhouse data.","USE_SAMPLE_DATA","(optional) Default Greenhouse Harvest key used when a user hasn't pasted their own AND the mock URL is unset. Leave blank for BYOK-only or mock-only setups.","GREENHOUSE_HARVEST_API_KEY","(optional) Default Anthropic key used when a signed-in user hasn't pasted their own. Leave blank to require per-user BYOK.","Postman mock URL (e.g. https://.mock.pstmn.io). Leave empty to skip the mock and hit real Greenhouse directly.","GREENHOUSE_MOCK_URL","(optional) HMAC secret for signed session cookies. Leave blank — the agent will generate one on first boot. Only set this if you want to keep users signed in across a re-deploy from scratch.","(optional) Fernet key for per-user credential encryption. Leave blank — the agent will generate one on first boot and persist it to its own database. Only set this if you're migrating from an older deploy and want to preserve existing encrypted user data.","CONNECTION_ENCRYPTION_KEY","20s","/healthz",20,"registry.astropods.ai/matt","https://assets.astropods.ai/avatars/agents/matt/pirate-parrot.jpg",{"_468":1494,"_470":1495,"_472":1496,"_474":475,"_476":1497,"_478":1498,"_480":1274,"_482":775,"_484":1499},[1467],{"_366":1465,"_367":1466},[1463],{"_61":1464,"_347":26},"Matthew Colozzo",54,10,{"_369":1468,"_371":1469,"_373":1470,"_375":1471,"_377":1472},"db40ea9a",{"_416":1483,"_418":1484,"_862":1485,"_422":1486,"_371":584},"---\ndescription: \"Replies to every user prompt with a semi-original sea shanty lyrics and pirate-themed emojis, powered solely by Anthropic.\"\ntags: [\"sea-shanty\", \"pirate\", \"for-fun\"]\nauthors: []\ncapabilities: [\"sea-shanty-generation\"]\nintegrations:\n - \"anthropic\"\n---\n\nReplies to every user prompt with the lyrics of an original sea shanty, themed around the user's message and dressed up with pirate emojis. Uses only Anthropic (Claude) for generation.\n",{"_379":1473,"_381":1474,"_385":1475,"_387":1476,"_389":1477},"2026-09-18T23:09:54Z","Replies to every user prompt with a semi-original sea shanty lyrics and pirate-themed emojis, powered solely by Anthropic.",[1480,1481,1482],[1479],"\nReplies to every user prompt with the lyrics of an original sea shanty, themed around the user's message and dressed up with pirate emojis. Uses only Anthropic (Claude) for generation.\n",[1478],{"_59":568,"_61":569,"_397":24},"sea-shanty-generation","sea-shanty","pirate","for-fun",{"_465":1493},{"_450":1490},{"_351":1489},{"_642":1487},{"_422":1488,"_426":645},[647],"private",{"_454":1491},{"_456":1492},[462],"registry.astropods.ai/matt/pirate-parrot:db40ea9a","#775a29","#fad89e","#9f6501","#86652d","#221a0b","#e0bf85","pdf-agent","https://assets.astropods.ai/avatars/agents/simon/pdf-agent.jpg",{"_468":1544,"_470":603,"_472":1545,"_474":475,"_476":477,"_478":479,"_480":481,"_482":483,"_484":607},[1507],{"_366":303,"_367":1048},[1506],{"_61":31,"_347":31},{"_369":1508,"_371":1509,"_373":1510,"_375":1511,"_377":1512},"75524c39",{"_416":1534,"_418":1535,"_958":1536,"_422":1537,"_371":1301},"---\ndescription: \"Fork of mayooear/ai-pdf-chatbot-langchain — upload PDFs and chat with them using LangChain and LangGraph, with documents embedded and stored in pgvector.\"\ntags: [rag, pdf, langchain, langgraph, document-qa, pgvector]\nauthors:\n - name: Simon Guerrier\n account: simwar\ncapabilities:\n - Ingest PDF documents and store them as vector embeddings\n - Answer questions about uploaded documents using semantic search\n - Maintain conversation history across multiple turns\nrepository: github:Simwar/ai-pdf-chatbot-langchain\nintegrations: [openai, postgres]\n---\n\n# AI PDF Chatbot\n\nUpload one or more PDF documents and ask questions about their content. The agent retrieves the most relevant passages using vector search and generates grounded answers via GPT-4o. This is a fork of the original [ai-pdf-chatbot-langchain](https://github.com/mayooear/ai-pdf-chatbot-langchain) by mayooear.\n\n## How it works\n\n1. **Ingest** — PDFs are parsed, chunked, embedded with `text-embedding-3-small`, and stored in a pgvector database.\n2. **Retrieve** — When you ask a question, the top-k most relevant chunks are retrieved using cosine similarity search.\n3. **Generate** — The retrieval graph passes the chunks as context to GPT-4o to produce a grounded answer with source citations.\n\n## Usage\n\nOpen the frontend at the agent URL, upload up to 5 PDFs (10 MB each), then type your question in the chat box. The agent will answer based on the content of your documents.\n\n## Limitations\n\n- PDF only — other file formats are not supported.\n- Max 5 files per upload, 10 MB per file.\n- The underlying LangGraph dev server is an in-memory store; ingested documents persist only for the lifetime of the running container unless a persistent volume is attached to the postgres knowledge store.\n",{"_379":1513,"_381":1514,"_559":1515,"_383":1516,"_385":1517,"_387":1518,"_389":1519},"2026-05-12T09:45:11Z","Fork of mayooear/ai-pdf-chatbot-langchain — upload PDFs and chat with them using LangChain and LangGraph, with documents embedded and stored in pgvector.",[1528,1529,1530,1531,1532,1533],[1527],{"_62":574,"_409":1526},[1523,1524,1525],"\n# AI PDF Chatbot\n\nUpload one or more PDF documents and ask questions about their content. The agent retrieves the most relevant passages using vector search and generates grounded answers via GPT-4o. This is a fork of the original [ai-pdf-chatbot-langchain](https://github.com/mayooear/ai-pdf-chatbot-langchain) by mayooear.\n\n## How it works\n\n1. **Ingest** — PDFs are parsed, chunked, embedded with `text-embedding-3-small`, and stored in a pgvector database.\n2. **Retrieve** — When you ask a question, the top-k most relevant chunks are retrieved using cosine similarity search.\n3. **Generate** — The retrieval graph passes the chunks as context to GPT-4o to produce a grounded answer with source citations.\n\n## Usage\n\nOpen the frontend at the agent URL, upload up to 5 PDFs (10 MB each), then type your question in the chat box. The agent will answer based on the content of your documents.\n\n## Limitations\n\n- PDF only — other file formats are not supported.\n- Max 5 files per upload, 10 MB per file.\n- The underlying LangGraph dev server is an in-memory store; ingested documents persist only for the lifetime of the running container unless a persistent volume is attached to the postgres knowledge store.\n",[1520,1521],{"_59":402,"_61":403,"_397":24},{"_59":965,"_61":1522,"_397":24},"PostgreSQL","Ingest PDF documents and store them as vector embeddings","Answer questions about uploaded documents using semantic search","Maintain conversation history across multiple turns","https://github.com/Simwar/ai-pdf-chatbot-langchain",{"_61":908,"_347":909},"rag","pdf","langchain","langgraph","document-qa","pgvector",{"_465":1542,"_450":1543},{"_450":1540},{"_1419":1539},{"_402":1538},{"_426":402},{"_1194":24,"_426":965},{"_452":1541},{"_463":1113},"registry.astropods.ai/simon/pdf-agent:75524c39",{"_452":24,"_454":242},"#8287cd","#5c68f2","payguard","https://assets.astropods.ai/avatars/agents/kraxx/payguard.jpg?v=1791548871",{"_468":1583,"_470":1584,"_472":1585,"_474":475,"_476":1586,"_478":1587,"_480":1588,"_482":1589,"_484":1590},[1553],{"_366":303,"_367":475},[1552],{"_61":49,"_347":49},{"_369":1554,"_371":1555,"_373":108,"_375":1556,"_377":1557},"fde7e6d0",{"_416":1558,"_422":1559,"_371":424},{"_387":108},"2026-10-09T13:56:16Z",{"_652":242,"_465":1562,"_420":1563,"_450":1564},{"_642":1560},{"_422":1561,"_426":645},[646],"registry.astropods.ai/5cd28510-ebca-4bec-94f6-ab5c8b59a5ef/payguard:fde7e6d0",[1565,1566,1567,1568,1569,1570],{"_452":24,"_454":242},{"_439":440,"_379":1581,"_61":1582,"_984":24},{"_439":440,"_379":1579,"_61":1580,"_445":24},{"_439":440,"_379":1577,"_61":1578,"_445":24},{"_439":440,"_379":1575,"_61":1576,"_445":24},{"_439":440,"_379":1573,"_61":1574,"_445":24},{"_439":440,"_379":1571,"_61":1572},"Sandbox business account merchant ID","PAYPAL_MERCHANT_ID","Separate Secret from the same PayPal Sandbox REST app","PAYPAL_CLIENT_SECRET","PayPal Sandbox REST app Client ID","PAYPAL_CLIENT_ID","Distinct random operator access token, 32 or more characters","OPERATOR_TOKEN","Random shopper access token, 32 or more characters","SHOPPER_TOKEN","Exact assigned HTTPS frontend origin, without trailing slash or path","APP_ORIGIN","#c4659d","#fa9ed4","#f435a6","#862d61","#220b19","#f6f4f5","#e699c6","#e085bb","open-deep-research","https://assets.astropods.ai/avatars/agents/simon/open-deep-research.jpg",{"_468":1664,"_470":1665,"_472":1666,"_474":475,"_476":1667,"_478":1314,"_480":875,"_482":1668,"_484":1669},[1598],{"_366":303,"_367":303},[1597],{"_61":31,"_347":31},{"_369":1599,"_371":1600,"_373":1601,"_375":1602,"_377":1603},"d62b072d",{"_416":1624,"_418":1625,"_420":1626,"_422":1627,"_371":584},"---\ndescription: \"Fork of langchain-ai/open_deep_research — autonomous deep research agent that runs parallel web searches across sub-agents and produces a comprehensive cited report.\"\ntags: [\"research\", \"deep-research\", \"report-generation\", \"web-search\", \"langgraph\"]\nauthors:\n - name: Simon Guerrier\n account: simwar\nrepository: github:Simwar/open_deep_research\ncapabilities:\n - \"Parallel web research using multiple concurrent sub-agents\"\n - \"Clarifying questions to scope research before starting\"\n - \"Comprehensive report generation with citations\"\n - \"Configurable search backends (Tavily, OpenAI, Anthropic native search)\"\n - \"Configurable research, summarization, and report models\"\nintegrations:\n - \"openai\"\n - \"anthropic\"\n - \"tavily\"\n---\n\nOpen Deep Research is a fully open-source autonomous research agent built on LangGraph. This is a fork of the original [open_deep_research](https://github.com/langchain-ai/open_deep_research) by LangChain AI. Give it any topic or question and it will scope the research (optionally asking clarifying questions), spin up parallel sub-agents to search the web, and synthesize findings into a well-structured, cited report.\n\n**Example prompts:**\n- \"What are the latest developments in fusion energy?\"\n- \"Compare the top vector database providers for production use\"\n- \"Write a competitive analysis of the API management market\"\n- \"Summarize the current state of AI regulation in the EU\"\n\n## Configuration\n\nAll settings are optional — sensible defaults are provided. Configure via deployment inputs:\n\n| Input | Default | Description |\n|-------|---------|-------------|\n| `TAVILY_API_KEY` | — | Required when `SEARCH_API=tavily` (default) |\n| `SEARCH_API` | `tavily` | Search backend: `tavily`, `openai`, `anthropic`, `none` |\n| `RESEARCH_MODEL` | `openai:gpt-4.1` | Model used by research sub-agents |\n| `FINAL_REPORT_MODEL` | `openai:gpt-4.1` | Model that writes the final report |\n| `SUMMARIZATION_MODEL` | `openai:gpt-4.1-mini` | Model for summarizing raw search results |\n| `COMPRESSION_MODEL` | `openai:gpt-4.1` | Model for compressing sub-agent findings |\n| `MAX_CONCURRENT_RESEARCH_UNITS` | `5` | Number of parallel sub-agents (1–20) |\n| `ALLOW_CLARIFICATION` | `true` | Ask a clarifying question before starting |\n\n## Getting a Tavily API Key\n\nTavily is the default search backend. To get an API key:\n\n1. Go to [app.tavily.com](https://app.tavily.com) and create a free account\n2. After signing in, navigate to **API Keys** in the dashboard\n3. Copy your API key and set it as the `TAVILY_API_KEY` deployment input\n\nThe free tier includes 1,000 searches/month. For higher usage, see [tavily.com/pricing](https://tavily.com/pricing).\n\n> **Tip:** If you don't want to use Tavily, set `SEARCH_API` to `openai` or `anthropic` to use those providers' built-in search tools instead — no Tavily key needed.\n\n## Limitations\n\n- Research quality depends on the configured search API and model; Tavily + GPT-4.1 is the recommended default\n- High concurrency settings may hit provider rate limits\n",{"_379":1604,"_381":1605,"_559":1606,"_383":1607,"_385":1608,"_387":1609,"_389":1610},"2026-05-12T09:27:13Z","Fork of langchain-ai/open_deep_research — autonomous deep research agent that runs parallel web searches across sub-agents and produces a comprehensive cited report.",[1621,1622,1623,1100,1531],[1620],{"_62":574,"_409":1619},[1614,1615,1616,1617,1618],"\nOpen Deep Research is a fully open-source autonomous research agent built on LangGraph. This is a fork of the original [open_deep_research](https://github.com/langchain-ai/open_deep_research) by LangChain AI. Give it any topic or question and it will scope the research (optionally asking clarifying questions), spin up parallel sub-agents to search the web, and synthesize findings into a well-structured, cited report.\n\n**Example prompts:**\n- \"What are the latest developments in fusion energy?\"\n- \"Compare the top vector database providers for production use\"\n- \"Write a competitive analysis of the API management market\"\n- \"Summarize the current state of AI regulation in the EU\"\n\n## Configuration\n\nAll settings are optional — sensible defaults are provided. Configure via deployment inputs:\n\n| Input | Default | Description |\n|-------|---------|-------------|\n| `TAVILY_API_KEY` | — | Required when `SEARCH_API=tavily` (default) |\n| `SEARCH_API` | `tavily` | Search backend: `tavily`, `openai`, `anthropic`, `none` |\n| `RESEARCH_MODEL` | `openai:gpt-4.1` | Model used by research sub-agents |\n| `FINAL_REPORT_MODEL` | `openai:gpt-4.1` | Model that writes the final report |\n| `SUMMARIZATION_MODEL` | `openai:gpt-4.1-mini` | Model for summarizing raw search results |\n| `COMPRESSION_MODEL` | `openai:gpt-4.1` | Model for compressing sub-agent findings |\n| `MAX_CONCURRENT_RESEARCH_UNITS` | `5` | Number of parallel sub-agents (1–20) |\n| `ALLOW_CLARIFICATION` | `true` | Ask a clarifying question before starting |\n\n## Getting a Tavily API Key\n\nTavily is the default search backend. To get an API key:\n\n1. Go to [app.tavily.com](https://app.tavily.com) and create a free account\n2. After signing in, navigate to **API Keys** in the dashboard\n3. Copy your API key and set it as the `TAVILY_API_KEY` deployment input\n\nThe free tier includes 1,000 searches/month. For higher usage, see [tavily.com/pricing](https://tavily.com/pricing).\n\n> **Tip:** If you don't want to use Tavily, set `SEARCH_API` to `openai` or `anthropic` to use those providers' built-in search tools instead — no Tavily key needed.\n\n## Limitations\n\n- Research quality depends on the configured search API and model; Tavily + GPT-4.1 is the recommended default\n- High concurrency settings may hit provider rate limits\n",[1611,1612,1613],{"_59":402,"_61":403,"_397":24},{"_59":568,"_61":569,"_397":24},{"_59":901,"_61":901,"_397":242},"Parallel web research using multiple concurrent sub-agents","Clarifying questions to scope research before starting","Comprehensive report generation with citations","Configurable search backends (Tavily, OpenAI, Anthropic native search)","Configurable research, summarization, and report models","https://github.com/Simwar/open_deep_research",{"_61":908,"_347":909},"research","deep-research","report-generation",{"_465":1663},{"_450":1660},{"_1630":1631,"_1632":1633,"_1634":1635,"_1636":1637,"_1638":1639,"_1640":1641,"_1642":1643,"_919":1644},{"_568":1628,"_402":1629},{"_426":568},{"_426":402},"ALLOW_CLARIFICATION",{"_439":1657,"_642":1658,"_379":1659,"_61":1630,"_984":24},"COMPRESSION_MODEL",{"_439":440,"_642":1651,"_379":1656,"_442":443,"_61":1632,"_984":24},"FINAL_REPORT_MODEL",{"_439":440,"_642":1651,"_379":1655,"_442":443,"_61":1634,"_984":24},"MAX_CONCURRENT_RESEARCH_UNITS",{"_439":671,"_642":1653,"_379":1654,"_442":443,"_61":1636,"_984":24},"RESEARCH_MODEL",{"_439":440,"_642":1651,"_379":1652,"_442":443,"_61":1638,"_984":24},"SEARCH_API",{"_439":440,"_642":901,"_379":1648,"_442":990,"_61":1640,"_984":24,"_991":1649},"SUMMARIZATION_MODEL",{"_439":440,"_642":1646,"_379":1647,"_442":443,"_61":1642,"_984":24},{"_439":440,"_379":1645,"_442":443,"_61":919,"_445":24},"Tavily API key for web search (used when SEARCH_API=tavily, which is the default)","openai:gpt-4.1-mini","Model for summarizing search results","Search backend to use: tavily | anthropic | openai | none",[901,568,402,1650],"none","openai:gpt-4.1","Model for research sub-agents, e.g. openai:gpt-4.1 or anthropic:claude-sonnet-4-5","5","Number of parallel research sub-agents (max: 20)","Model for writing the final report","Model for compressing sub-agent findings","boolean","true","Ask a clarifying question before starting research",{"_454":1661},{"_456":1662},[462],"registry.astropods.ai/simon/open-deep-research:d62b072d","#c28e83","#eeb6aa","#e17a64","#863c2d","#e4a89a","#e09585","on-this-day-in-art","https://assets.astropods.ai/avatars/agents/alohajha/on-this-day-in-art.jpg",{"_468":1711,"_470":1584,"_472":1712,"_474":475,"_476":1586,"_478":1587,"_480":1588,"_482":1713,"_484":1590},[1678],{"_366":1677,"_367":1130},[1676],{"_61":145,"_347":47},39,{"_369":1679,"_371":1680,"_373":1681,"_375":1682,"_377":1683},"ff19e734",{"_416":1698,"_418":1699,"_862":1700,"_422":1701,"_371":584},"---\ndescription: \"Art discovery agent — one Met Museum artwork per request, with Chillomena Punk commentary. Say 'today', a date, or search by keyword.\"\ntags: [art, met-museum, comedy, discovery]\nauthors: []\ncapabilities: [tool-use, mcp]\nintegrations:\n - anthropic\n - gemini\n---\n\n# On This Day in Art\n\nAn art discovery agent powered by the Metropolitan Museum of Art's public API. Surfaces one artwork connected to today's date — or any date the user picks — and presents it with a short commentary in the voice of **Chillomena Punk**: a deadpan, confidently misinformed art critic inspired by Philomena Cunk.\n\n## Three modes\n\n| Mode | How to trigger |\n|------|---------------|\n| **On This Day** | \"today\", \"what's on today?\", \"show me today's art\" |\n| **Pick a Day** | Any date: \"June 9\", \"March 3rd\", \"12/25\", \"1/15/1990\" |\n| **Search** | Keywords, artists, themes: \"show me a Rembrandt\", \"something with flowers\", \"Japanese woodblock prints\" |\n\n## Tool pipeline\n\nEach artwork request runs three tools in sequence:\n\n1. `get_artworks_for_today` or `search_artworks` — finds the artwork by date (with three-tier fallback) or keyword\n2. `get_artwork_detail` — fetches full metadata for the selected artwork\n3. `get_gemini_summary` — generates ~150-word Chillomena Punk commentary via Gemini 2.5 Flash\n",{"_379":1684,"_381":1685,"_385":1686,"_387":1687,"_389":1688},"2026-06-10T16:33:04Z","Art discovery agent — one Met Museum artwork per request, with Chillomena Punk commentary. Say 'today', a date, or search by keyword.",[1694,1695,1696,1697],[1692,1693],"\n# On This Day in Art\n\nAn art discovery agent powered by the Metropolitan Museum of Art's public API. Surfaces one artwork connected to today's date — or any date the user picks — and presents it with a short commentary in the voice of **Chillomena Punk**: a deadpan, confidently misinformed art critic inspired by Philomena Cunk.\n\n## Three modes\n\n| Mode | How to trigger |\n|------|---------------|\n| **On This Day** | \"today\", \"what's on today?\", \"show me today's art\" |\n| **Pick a Day** | Any date: \"June 9\", \"March 3rd\", \"12/25\", \"1/15/1990\" |\n| **Search** | Keywords, artists, themes: \"show me a Rembrandt\", \"something with flowers\", \"Japanese woodblock prints\" |\n\n## Tool pipeline\n\nEach artwork request runs three tools in sequence:\n\n1. `get_artworks_for_today` or `search_artworks` — finds the artwork by date (with three-tier fallback) or keyword\n2. `get_artwork_detail` — fetches full metadata for the selected artwork\n3. `get_gemini_summary` — generates ~150-word Chillomena Punk commentary via Gemini 2.5 Flash\n",[1689,1690],{"_59":568,"_61":569,"_397":24},{"_59":1691,"_61":1691,"_397":242},"gemini","tool-use","mcp","art","met-museum","comedy","discovery",{"_465":1706,"_420":1707},{"_450":1703},{},{"_568":1702},{"_426":568},{"_454":1704},{"_456":1705},[462],"registry.astropods.ai/7553f065-7ec0-4a0f-aa70-0298b29e00c7/on-this-day-in-art:ff19e734",[1708],{"_439":440,"_379":1709,"_61":1710,"_445":24},"Google Gemini API key — used by the MCP server for Chillomena Punk commentary","GEMINI_API_KEY","#a93a7b","#e10285","#b6a7d7","nextgig","https://assets.astropods.ai/avatars/agents/astro-ai/nextgig.jpg?v=1788169709",{"_468":1778,"_470":1779,"_472":1780,"_474":475,"_476":1781,"_478":1782,"_480":1783,"_482":1784,"_484":1785},[1722],{"_366":303,"_367":1721},[1720],{"_61":31,"_347":31},7,{"_369":1723,"_371":1724,"_373":1725,"_375":1726,"_377":1727},"3fa4c1e7",{"_416":1758,"_418":1759,"_958":1760,"_862":1761,"_422":1762,"_371":584},"---\ndescription: \"A personal job-hunting agent: turns your LinkedIn/CV into a skill matrix, emails verified new job matches, and tracks what you applied to.\"\ntags:\n - jobs\n - job-search\n - linkedin\n - cv\n - email-digest\n - notifications\n - web-search\n - application-tracking\n - personal-agent\nauthors:\n - name: \"Simon Guerrier\"\n account: \"simon\"\nrepository: \"github:Simwar/nextgig\"\ncapabilities:\n - \"Extract a structured skill matrix from a pasted LinkedIn profile or uploaded PDF/CV\"\n - \"Capture target location and remote (remote/hybrid/onsite) preferences\"\n - \"Search the web for currently-open job postings matching the candidate\"\n - \"Open each posting to check it is really that role and still live before presenting it\"\n - \"Email daily or weekly job-match digests via Resend, de-duplicated across runs\"\n - \"Track applications and their status, and never re-suggest a role you applied to\"\n - \"Search and preview matches on demand from the chat UI\"\nintegrations:\n - \"Astro AI Gateway\"\n - \"Resend\"\n - \"PostgreSQL\"\n---\n\n

\n $\"NextGig\"$ \n

NextGig

Hand it your CV once. It runs the job hunt from there.

\n\nJob hunting is the same loop over and over: run the same searches, re-read listings you've already seen, click through to a posting that turned out to be filled last week, then lose track of what you actually applied to.\n\nNextGig runs that loop for you. Give it your LinkedIn profile or CV once and it emails you a short digest of **new, still-open** roles that match your skills and location — nothing you've been shown before. Tell it when you apply, and it keeps the list.\n\n

\n $\"The$ \n

\n $\"NextGig\"$ \n

NextGig

Hand it your CV once. It runs the job hunt from there.

\n $\"The$ \n

\n\n> **Personal (single-user) agent.** Deploy it by yourself, for yourself, and keep the deployment private. There is no authentication, email settings are shared per deployment, and all model and email usage bills to your account. See [`docs/single-vs-multi-user.md`](docs/single-vs-multi-user.md) for what running it for other people would take.\n\n## Overview\n\nA chat agent with its own small web frontend. Onboarding is conversational and takes about a minute:\n\n1. **Profile → skill matrix.** Upload your LinkedIn profile as a PDF (LinkedIn → More → Save to PDF) or your CV, or just paste the text. It extracts your headline, target job titles, and a skill list with proficiency levels. PDFs are parsed to text server-side — LinkedIn is never scraped.\n2. **Location & preferences.** Where you want to work, and whether you want remote, hybrid, onsite, or any.\n3. **Notifications.** Your email and a cadence — daily or weekly. Optional; chat works without it.\n\nFrom then on a built-in scheduler does the hunting:\n\n- **Finds matches** with the gateway's server-side web search, biased toward individual postings on applicant-tracking systems (Greenhouse, Lever, Ashby) and company career pages rather than aggregator homepages.\n- **Checks every link before you see it** — it opens each candidate posting and drops the dead ones (404 / expired / filled). If a page can't prove it's really that role — many boards answer an unknown job id with their generic careers page, and some render entirely in the browser — the role is labelled *unconfirmed* rather than quietly presented as verified.\n- **De-duplicates** against everything already sent, so a digest never repeats a role.\n- **Emails only what's new**, and sends nothing at all when there's nothing new.\n- **Remembers your applications.** Applied roles never come back in a digest.\n\nNothing here needs an API key of your own: the model *and* the web search run through the **Astro AI Gateway** on a managed per-tenant credential. State lives in a **Postgres store deployed alongside the agent**, so your profile, applications, and history survive restarts and redeploys.\n\n## Usage\n\n- **Onboard** — open the frontend, drag your CV/LinkedIn PDF anywhere onto the page, and answer the two follow-up questions.\n- **Get matches now** — *\"Send me matches now\"* runs a search and emails the digest. *\"What roles are out there for me today?\"* discusses them in chat instead.\n- **Adjust anything** — *\"Change my location to Berlin, remote only\"*, *\"Switch me to weekly emails\"*.\n- **Track applications** — *\"I applied to #2\"*, *\"I have an interview with Wolt on Tuesday\"*, *\"Acme rejected me\"*. Then *\"What have I applied to?\"* for a dated list with links and statuses (applied, interviewing, offer, rejected, withdrawn).\n- **Enable email** — gear icon, top right → paste a free [Resend](https://resend.com) API key → **Save & send test**. Or set `RESEND_API_KEY` as a deployment secret, which takes precedence.\n\n## Deploy your own\n\n```bash\nast login # the gateway and the store are account-scoped\nbun install\nast project start # local dev: agent on :3200, plus a Postgres sidecar\n\nast push --visibility public\nast deploy nextgig --wait\n```\n\nA Resend key is the only key involved, it is optional, and it can be pasted into the app instead of deployed as a secret.\n\n### Configuration\n\n| Variable | Purpose | Required |\n|---|---|---|\n| `ASTRO_GATEWAY_URL` / `ASTRO_GATEWAY_API_KEY` / `MODEL_DEFAULT` | Model + server-side web search, from the `models.default` gateway block in `astropods.yml` | Injected |\n| `POSTGRES_HOST` / `PORT` / `USER` / `PASSWORD` / `DB` | The `knowledge.db` Postgres store deployed with the agent | Injected |\n| `RESEND_API_KEY` | Email delivery — or set it in-app via the Settings panel | For email digests |\n| `RESEND_FROM` | Sender address; defaults to `onboarding@resend.dev` | No |\n\n## How it works\n\n| Piece | Choice |\n|---|---|\n| Model | `gpt-5-6-luna` through the Astro AI Gateway (`models.default.provider: gateway`) — swap it in `astropods.yml` |\n| Job search | The gateway's provider-executed `web_search`, so no search API key |\n| Link checking | A client-side `web_fetch` tool that opens one URL and reports whether the page backs up the posting |\n| Storage | Postgres knowledge store — profile, applications, de-dup history, chat memory |\n| Scheduling | In-process `node-cron`, hourly tick; email sent deterministically in code, not by a tool call |\n| Interface | Custom chat + PDF-upload frontend on port 80; no messaging sidecar |\n\nThe gateway wiring has sharp edges worth reading before you fork this — [`docs/ai-gateway.md`](docs/ai-gateway.md) covers why model calls must set `store: false`, why web search runs with `externalWebAccess: false`, and why replies are non-streaming.\n\n## Limitations\n\n- **Personal use only** — no authentication or per-user accounts, and email settings are shared per deployment. Keep the URL private.\n- **Search freshness varies.** Web search reads a cached index rather than the live web, so fast-moving markets show gaps and a run can legitimately return nothing new.\n- **Verification has a ceiling.** Postings on browser-rendered boards (Ashby, Workday) and behind bot walls (LinkedIn, Indeed) cannot be confirmed server-side — they are shown as unconfirmed, never silently as verified.\n- **Scanned or image-only PDFs** cannot be parsed; use a text export or paste your profile.\n- **Replies arrive in one piece.** A search turn can take up to a minute, so the UI streams progress (\"Searching the web…\", \"Opening jobs.ashbyhq.com…\") rather than the answer token by token.\n",[1735,1736,1737],{"_59":1740,"_61":1741,"_397":242},{"_59":1738,"_61":1739,"_397":242},{"_59":965,"_61":1522,"_397":24},"resend","Resend","astro ai gateway","Astro AI Gateway","Extract a structured skill matrix from a pasted LinkedIn profile or uploaded PDF/CV","Capture target location and remote (remote/hybrid/onsite) preferences","Search the web for currently-open job postings matching the candidate","Open each posting to check it is really that role and still live before presenting it","Email daily or weekly job-match digests via Resend, de-duplicated across runs","Track applications and their status, and never re-suggest a role you applied to","Search and preview matches on demand from the chat UI","https://github.com/Simwar/nextgig",{"_61":908,"_347":31},"jobs","job-search","linkedin","cv","email-digest","application-tracking","personal-agent",{"_465":1769,"_420":1770,"_450":1771},{"_450":1767},{"_1419":1766},{},{"_642":1763},{"_422":1764,"_426":645},[1765],"gpt-5-6-luna",{"_426":965},{"_452":1768},{"_463":1113},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/nextgig:3fa4c1e7",[1772,1773],{"_452":24},{"_439":440,"_379":1776,"_61":1777,"_984":24,"_445":24},{"_439":440,"_379":1774,"_61":1775,"_984":24,"_445":242},"From address, e.g. \"NextGig \". Defaults to onboarding@resend.dev (verify a domain in Resend to email arbitrary recipients).","RESEND_FROM","Resend API key (re_…). Get one free at resend.com. Optional — can be set in-app instead.","RESEND_API_KEY","#a878b1","#dfb0e8","#bc5bce","#782d86","#1f0b22","#f6f4f6","#db9fe0","#cf89dc","muster","https://assets.astropods.ai/avatars/agents/rabbah/muster.jpg",{"_468":1834,"_470":1835,"_472":1836,"_474":475,"_476":1211,"_478":1212,"_480":875,"_482":1837,"_484":1838},[1793],{"_366":303,"_367":303},[1792],{"_61":192,"_347":37},{"_369":1794,"_371":1795,"_373":1796,"_375":1797,"_377":1798},"59e65410",{"_416":1816,"_418":1817,"_422":1818,"_371":584},"---\ndescription: Minimal dispatcher agent that runs bundled markdown skill files on deploy-time cron schedules and posts results to Slack.\ntags:\n - skills\n - cron\n - mastra\n - anthropic\n - slack\nauthors:\n - name: Rodric Rabbah\n account: rabbah\ncapabilities:\n - Loads markdown skill files from a bundled /skills directory at boot\n - Routes chat messages to skills via a run_skill tool or explicit /skill-name slash dispatch\n - Runs skills on cron schedules declared at deploy time via MUSTER_SCHEDULES\n - Invokes the Mastra agent in-process and posts proactive AgentResponses to a Slack channel for scheduled runs\n - Logs scheduled run output to stdout for visibility without persistence\n - Sweeps stale cron-fired conversations from Mastra memory hourly with a 24h TTL\nrepository: github:rabbah/muster\nintegrations:\n - Anthropic\n - Slack\n - Mastra\n---\n\n## Overview\n\nRuns skills on demand (via chat) or on a schedule.\nSkills live as markdown files committed to the repo under `skills/`, \nand cron schedules are declared as a single deploy-time input.\n\n## Usage\n\n### Adding skills\n\nDrop markdown files into `skills/`. The filename (minus extension) becomes\nthe skill name. Accepted extensions: `.md`, `.markdown`, `.txt`. The first\nnon-empty line (stripped of leading `#`s) is shown to the LLM as the skill's\ndescription. Rebuild and redeploy for changes to take effect.\n\n### Chatting with the agent\n\nIn any Slack channel where the bot is installed, or via any other configured\nmessaging adapter:\n\n- Ask any question. The agent picks the best matching skill, calls\n `run_skill` to load it, and follows its instructions.\n- Prefix a message with `/` to dispatch directly to a specific\n skill, skipping LLM routing. Example: `/example show me what you do`.\n\n### Scheduling skills\n\nSet `MUSTER_SCHEDULES` to a comma-separated list of `:`\nentries — for example `digest:0 9 * * *,sweep:0,15,30,45 * * * *`. Each\nentry's cron part is a standard 5-field expression. Invalid entries (bad\ncron, unknown skill) are logged and skipped at boot.\n\nBy default `MUSTER_SCHEDULES=example:* * * * *` so a fresh install fires\nthe bundled example skill every minute. Override (or set to an empty\nstring) before wiring up real schedules.\n\n### Posting cron results to Slack\n\nSet `SLACK_CHANNEL` to a Slack channel ID (e.g. `C0123456789`) — **not** a\nchannel name. The bot must be a member of the channel; run\n`/invite @` first, or use a DM channel (`D…` IDs always work).\nSlack bot credentials are managed by the platform's Slack adapter, not\ndeclared here.\n\n### Inputs\n\n| Input | Required | Description |\n|---|---|---|\n| `MUSTER_SCHEDULES` | no | Comma-separated `:` entries. Default: `example:* * * * *`. |\n| `SLACK_CHANNEL` | no | Slack channel ID to post cron results to. Requires the platform's Slack adapter. |\n\n## Limitations\n\n- **Skills are bundled at build time.** Adding or editing a skill requires\n rebuilding the image and redeploying.\n- **Scheduler runs in-process; single replica only.** With multiple agent\n replicas the same cron job fires on each.\n- **No persistence of run history.** Cron results go to stdout and\n (optionally) Slack; nothing is stored. Inspect past runs via your log\n pipeline.\n- **Mastra memory is in-process.** Each cron fire creates a fresh thread\n in `LibSQLStore({ url: ':memory:' })`; an hourly sweep deletes threads\n older than 24h, but they don't survive a process restart anyway.\n- **Bot must be in the target Slack channel.** Posts to channels the bot\n isn't a member of are silently dropped (`not_in_channel`).\n",{"_379":1799,"_381":1800,"_559":1801,"_383":1802,"_385":1803,"_387":1804,"_389":1805},"2026-05-21T16:16:19Z","Minimal dispatcher agent that runs bundled markdown skill files on deploy-time cron schedules and posts results to Slack.",[1098,1099,579,568,458],[1815],{"_62":574,"_409":1814},[1809,1810,1811,1812,1813,1093],"\n## Overview\n\nRuns skills on demand (via chat) or on a schedule.\nSkills live as markdown files committed to the repo under `skills/`, \nand cron schedules are declared as a single deploy-time input.\n\n## Usage\n\n### Adding skills\n\nDrop markdown files into `skills/`. The filename (minus extension) becomes\nthe skill name. Accepted extensions: `.md`, `.markdown`, `.txt`. The first\nnon-empty line (stripped of leading `#`s) is shown to the LLM as the skill's\ndescription. Rebuild and redeploy for changes to take effect.\n\n### Chatting with the agent\n\nIn any Slack channel where the bot is installed, or via any other configured\nmessaging adapter:\n\n- Ask any question. The agent picks the best matching skill, calls\n `run_skill` to load it, and follows its instructions.\n- Prefix a message with `/` to dispatch directly to a specific\n skill, skipping LLM routing. Example: `/example show me what you do`.\n\n### Scheduling skills\n\nSet `MUSTER_SCHEDULES` to a comma-separated list of `:`\nentries — for example `digest:0 9 * * *,sweep:0,15,30,45 * * * *`. Each\nentry's cron part is a standard 5-field expression. Invalid entries (bad\ncron, unknown skill) are logged and skipped at boot.\n\nBy default `MUSTER_SCHEDULES=example:* * * * *` so a fresh install fires\nthe bundled example skill every minute. Override (or set to an empty\nstring) before wiring up real schedules.\n\n### Posting cron results to Slack\n\nSet `SLACK_CHANNEL` to a Slack channel ID (e.g. `C0123456789`) — **not** a\nchannel name. The bot must be a member of the channel; run\n`/invite @` first, or use a DM channel (`D…` IDs always work).\nSlack bot credentials are managed by the platform's Slack adapter, not\ndeclared here.\n\n### Inputs\n\n| Input | Required | Description |\n|---|---|---|\n| `MUSTER_SCHEDULES` | no | Comma-separated `:` entries. Default: `example:* * * * *`. |\n| `SLACK_CHANNEL` | no | Slack channel ID to post cron results to. Requires the platform's Slack adapter. |\n\n## Limitations\n\n- **Skills are bundled at build time.** Adding or editing a skill requires\n rebuilding the image and redeploying.\n- **Scheduler runs in-process; single replica only.** With multiple agent\n replicas the same cron job fires on each.\n- **No persistence of run history.** Cron results go to stdout and\n (optionally) Slack; nothing is stored. Inspect past runs via your log\n pipeline.\n- **Mastra memory is in-process.** Each cron fire creates a fresh thread\n in `LibSQLStore({ url: ':memory:' })`; an hourly sweep deletes threads\n older than 24h, but they don't survive a process restart anyway.\n- **Bot must be in the target Slack channel.** Posts to channels the bot\n isn't a member of are silently dropped (`not_in_channel`).\n",[1806,1807,1808],{"_59":568,"_61":569,"_397":24},{"_59":458,"_61":1083,"_397":24},{"_59":579,"_61":1082,"_397":242},"Loads markdown skill files from a bundled /skills directory at boot","Routes chat messages to skills via a run_skill tool or explicit /skill-name slash dispatch","Runs skills on cron schedules declared at deploy time via MUSTER_SCHEDULES","Invokes the Mastra agent in-process and posts proactive AgentResponses to a Slack channel for schedu","Logs scheduled run output to stdout for visibility without persistence","https://github.com/rabbah/muster",{"_61":192,"_347":37},{"_465":1824,"_420":1825,"_450":1826},{"_450":1820},{"_568":1819},{"_426":568},{"_454":1821},{"_456":1822,"_458":1823},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/e9945c0f-c341-421f-8680-473e88c1807c/muster:59e65410",[1827,1828],{"_454":24},{"_439":440,"_642":1831,"_379":1832,"_61":1833,"_984":24},{"_439":440,"_379":1829,"_61":1830,"_984":24},"Slack channel ID (preferred) or `#name` to post cron run results to. Requires the `slack` messaging adapter to be enabled (Slack bot credentials are managed by the platform, not declared here).","SLACK_CHANNEL","example:* * * * *","Comma-separated cron schedules `:` (e.g. `digest:0 9 * * *,sweep:*/15 * * * *`). The skill name must match a file in /skills. Default fires the bundled `example` skill every minute so a fresh install has visible activity out of the box.","MUSTER_SCHEDULES","#d5b0b1","#ecacaf","#e89da0","#e6999c","#e08588","mindcraft","registry.astropods.ai/astro-ai","https://assets.astropods.ai/avatars/agents/astro-ai/mindcraft.jpg",{"_468":1899,"_470":1900,"_472":1899,"_474":475,"_476":1901,"_478":1902,"_480":1903,"_482":1904,"_484":1905},[1848],{"_366":303,"_367":1847},[1846],{"_61":31,"_347":31},15,{"_369":1849,"_371":1850,"_373":1851,"_375":1852,"_377":1853},"d1b5cb37",{"_416":1872,"_418":1873,"_420":1874,"_862":1875,"_371":1301},"---\ndescription: \"Claude-powered Minecraft bot that fights autonomously in a live battle royale arena. Give it a strategy and watch it play.\"\ntags:\n - minecraft\n - gaming\n - battle-royale\n - claude\n - ai-agent\nauthors:\n - name: Astropods\n account: Astropods\ncapabilities:\n - Navigates a Minecraft world and loots chests autonomously\n - Crafts weapons and armor from gathered resources\n - Hunts and fights other bots using configurable strategies\n - Adapts playstyle based on a plain-English strategy prompt\n - Registers and competes in live arena heats automatically\nintegrations:\n - anthropic\n---\n\n![Minecraft Arena](https://minecraft.wiki/images/1.21_banner.jpg)\n\n# Astro Arena Bot\n\nAn AI-powered Minecraft bot that fights autonomously in a live battle royale arena, powered by Claude. Drop it in, give it a strategy, and watch it compete.\n\nThe arena is running live — connect and spectate at any time.\n\n| Service | Address |\n|---------|---------|\n| Minecraft server | `srv.minecraft.astropod.ai:25565` |\n| Arena controller (UI + leaderboard) | [ctl.minecraft.astropod.ai](https://ctl.minecraft.astropod.ai) |\n\n## How to spectate\n\nYou don't need a paid Minecraft account — the server runs in offline mode.\n\n1. Download the [Minecraft Launcher](https://www.minecraft.net/en-us/download)\n2. Create a **1.21.6** installation under **Installations → New installation**\n3. Launch the game and go to **Multiplayer → Add Server**\n4. Enter `srv.minecraft.astropod.ai` and join\n\nYou'll spawn as a spectator — fly around and watch the bots fight. No account purchase required.\n\n> Don't have Minecraft? Watch the live leaderboard and heat status at **[ctl.minecraft.astropod.ai](https://ctl.minecraft.astropod.ai)** instead.\n\n## Configuring your bot\n\n### Strategy\n\nUse `BOT_STRATEGY` to tell the bot how to play. Write any plain-English instruction:\n\n| Example | Style |\n|---------|-------|\n| _(leave empty)_ | Balanced: loot → craft → hunt |\n| `Rush the nearest enemy immediately. Never retreat.` | Aggressive |\n| `Loot and fully gear up before engaging anyone.` | Careful crafter |\n| `Stay near the world border edge and let others fight first.` | Patient opportunist |\n\n### Model\n\nUse `BOT_MODEL` to trade off speed vs. capability:\n\n| Model | Speed | Cost |\n|-------|-------|------|\n| `claude-haiku-4-5-20251001` | Fastest | Cheapest |\n| `claude-sonnet-4-6` _(default)_ | Balanced | Medium |\n| `claude-opus-4-6` | Strongest | Most expensive |\n\n## Arena rules\n\n- **Flat plains world** — all resources come from loot chests inside buildings\n- **Central Tower** `(0,0)` — best loot (iron armor + axe)\n- **Outposts** at `(±42, 0)` and `(0, ±42)` — medium loot (sword + leather armor)\n- **World border** shrinks toward the center over 5 minutes — stay inside or take damage\n- **No respawning** — last bot standing wins the heat\n\n## Heat lifecycle\n\n1. **Idle** — waiting for 2+ bots to connect\n2. **Starting** — 30-second countdown (bots are invincible)\n3. **Running** — fight! Border shrinks from 200×200 to 30×30\n4. **Results** — winner and leaderboard announced\n5. **Resetting** — 2-minute break, then the next heat starts automatically\n",{"_379":1854,"_381":1855,"_559":1856,"_385":1857,"_387":1858,"_389":1859},"2026-06-10T07:57:37Z","Claude-powered Minecraft bot that fights autonomously in a live battle royale arena. Give it a strategy and watch it play.",[1867,1868,1869,1870,1871],[1866],[1861,1862,1863,1864,1865],"\n![Minecraft Arena](https://minecraft.wiki/images/1.21_banner.jpg)\n\n# Astro Arena Bot\n\nAn AI-powered Minecraft bot that fights autonomously in a live battle royale arena, powered by Claude. Drop it in, give it a strategy, and watch it compete.\n\nThe arena is running live — connect and spectate at any time.\n\n| Service | Address |\n|---------|---------|\n| Minecraft server | `srv.minecraft.astropod.ai:25565` |\n| Arena controller (UI + leaderboard) | [ctl.minecraft.astropod.ai](https://ctl.minecraft.astropod.ai) |\n\n## How to spectate\n\nYou don't need a paid Minecraft account — the server runs in offline mode.\n\n1. Download the [Minecraft Launcher](https://www.minecraft.net/en-us/download)\n2. Create a **1.21.6** installation under **Installations → New installation**\n3. Launch the game and go to **Multiplayer → Add Server**\n4. Enter `srv.minecraft.astropod.ai` and join\n\nYou'll spawn as a spectator — fly around and watch the bots fight. No account purchase required.\n\n> Don't have Minecraft? Watch the live leaderboard and heat status at **[ctl.minecraft.astropod.ai](https://ctl.minecraft.astropod.ai)** instead.\n\n## Configuring your bot\n\n### Strategy\n\nUse `BOT_STRATEGY` to tell the bot how to play. Write any plain-English instruction:\n\n| Example | Style |\n|---------|-------|\n| _(leave empty)_ | Balanced: loot → craft → hunt |\n| `Rush the nearest enemy immediately. Never retreat.` | Aggressive |\n| `Loot and fully gear up before engaging anyone.` | Careful crafter |\n| `Stay near the world border edge and let others fight first.` | Patient opportunist |\n\n### Model\n\nUse `BOT_MODEL` to trade off speed vs. capability:\n\n| Model | Speed | Cost |\n|-------|-------|------|\n| `claude-haiku-4-5-20251001` | Fastest | Cheapest |\n| `claude-sonnet-4-6` _(default)_ | Balanced | Medium |\n| `claude-opus-4-6` | Strongest | Most expensive |\n\n## Arena rules\n\n- **Flat plains world** — all resources come from loot chests inside buildings\n- **Central Tower** `(0,0)` — best loot (iron armor + axe)\n- **Outposts** at `(±42, 0)` and `(0, ±42)` — medium loot (sword + leather armor)\n- **World border** shrinks toward the center over 5 minutes — stay inside or take damage\n- **No respawning** — last bot standing wins the heat\n\n## Heat lifecycle\n\n1. **Idle** — waiting for 2+ bots to connect\n2. **Starting** — 30-second countdown (bots are invincible)\n3. **Running** — fight! Border shrinks from 200×200 to 30×30\n4. **Results** — winner and leaderboard announced\n5. **Resetting** — 2-minute break, then the next heat starts automatically\n",[1860],{"_59":568,"_61":569,"_397":24},"Navigates a Minecraft world and loots chests autonomously","Crafts weapons and armor from gathered resources","Hunts and fights other bots using configurable strategies","Adapts playstyle based on a plain-English strategy prompt","Registers and competes in live arena heats automatically",{"_61":272,"_347":27},"minecraft","gaming","battle-royale","claude","ai-agent",{"_1427":24,"_465":1898},{"_586":1897},{"_1876":1877,"_1878":1879,"_1880":1881,"_1882":1883,"_1884":1885,"_1886":1887},{"_351":1489},"ARENA_CONTROLLER_URL",{"_439":440,"_642":1895,"_379":1896,"_61":1876},"BOT_MODEL",{"_439":440,"_642":647,"_379":1894,"_61":1878},"BOT_NAME",{"_439":440,"_379":1893,"_61":1880},"BOT_STRATEGY",{"_439":440,"_379":1892,"_442":983,"_61":1882},"MINECRAFT_HOST",{"_439":440,"_642":1890,"_379":1891,"_61":1884},"MINECRAFT_PORT",{"_439":440,"_642":1888,"_379":1889,"_61":1886},"25565","Minecraft server port","srv.minecraft.astroids.ai","Minecraft server host (use host.docker.internal to reach localhost from inside Docker)","Free-text strategy injected into the bot's system prompt. Describe how the bot should play (e.g. 'Rush enemies immediately, never retreat.' or 'Loot first, engage only when fully armed.'). Leave empty for the balanced default.","Bot name shown in-game — must be unique per bot and can't contain space or special characters","Claude model to use (e.g. claude-haiku-4-5-20251001 for fast/cheap, claude-sonnet-4-6 for balanced, claude-opus-4-6 for strongest)","https://ctl.minecraft.astroids.ai/","Arena Controller base URL — bot registers here on startup","sh ./start.sh","registry.astropods.ai/astro-ai/mindcraft:d1b5cb37","#292929","#cccccc","#595959","#171717","#f5f5f5","#b999e6","#b3b3b3","memory-box","https://assets.astropods.ai/avatars/agents/chrisjpatty/memory-box.jpg",{"_468":1970,"_470":1971,"_472":1972,"_474":475,"_476":1973,"_478":1974,"_480":875,"_482":1975,"_484":1976},[1914],{"_366":303,"_367":1913},[1912],{"_61":167,"_347":45},36,{"_369":1915,"_371":1916,"_373":1917,"_375":1918,"_377":1919},"ebf00db6",{"_416":1938,"_418":1939,"_958":1940,"_862":1941,"_422":1942,"_371":584},"---\ndescription: \"Your personal knowledge base, powered by AI. Save anything and find it again through natural conversation.\"\ntags:\n - storage\n - semantic-search\n - mcp\ncapabilities:\n - Ingest arbitrary content\n - Search content with semantic search and natural language\n - Ingest and search via API or a chat UI\n - Import GitHub stars, and Twitter/X bookmarks\n - Connect via MCP to AI chat clients\nintegrations:\n - \"anthropic\"\n - \"github\"\n - \"twitter\"\n - \"ollama\"\n---\n\n

\n $\"Memory$ \n

\n\n

memory-box

\n $\"Memory$ \n

\n\n

memory-box

\n\nHow many times have you searched for the exact wording of something you know you have saved and it never comes up? Search is fundamentally broken in so many services as it competes with the goal of keeping you running more and more searches. Memory Box is designed to let you pull all your saved content from other services into one place and use search tools that just work.\n\n---\n\n## What It Does\n\nMemoryBox is a private AI agent that stores and retrieves your content. Send it a link, paste in a note, upload an image — it automatically classifies, tags, and indexes everything so you can search by meaning, not just keywords.\n\nAsk it \"what did I save last week about CSS animations?\" and it finds the relevant articles, notes, and code repos across your entire collection.\n\nOut of the box you get simple API access protected by bearer tokens and an MCP server that you can plug into chat clients like Claude or ChatGPT and get a one-click install.\n\n## Features\n\n### Save Anything\n\nDrop in any content and MemoryBox handles the rest:\n\n- **Links & Articles** — saves the full content, not just the URL\n- **Notes & Text** — quick thoughts, ideas, quotes, reminders\n- **Images** — analyzes visual content so you can search by what's in the picture\n- **Tweets** — preserves the post with author, metrics, and context\n- **GitHub Repos** — captures README, metadata, and key details\n\n### Natural Language Search\n\nEvery item is automatically:\n\n- Ingested with vector embeddings for fast semantic search\n- Classified and tagged into a categories and topics\n- Deduplicated so you never save the same thing twice\n\n### Bulk Import\n\nBring in content from the platforms you already use:\n\n- **GitHub** — import your starred repositories\n- **Twitter/X** — import your bookmarks, organized by folder\n- **More to come...**\n\n### Dashboard\n\nA web interface for browsing, searching, and managing your memories. Upload files, review imports, generate API tokens, and explore your collection visually.\n\n![Screenshot of the Memory Box search interface](https://i.ibb.co/KcScSJcW/Screenshot-2026-04-14-at-1-30-00-PM.png)\n\n### API & MCP Access\n\nIntegrate MemoryBox into your workflow:\n\n- **REST API** for programmatic ingestion and search\n- **MCP server** so AI assistants like Claude can read and write to your memory store directly\n",[1925,1926,1927,1928],{"_59":568,"_61":569,"_397":24},{"_59":408,"_61":1337,"_397":24},{"_59":1930,"_61":1930,"_397":242},{"_59":1929,"_61":1929,"_397":242},"ollama","twitter","Ingest arbitrary content","Search content with semantic search and natural language","Ingest and search via API or a chat UI","Import GitHub stars, and Twitter/X bookmarks","Connect via MCP to AI chat clients","storage","semantic-search",{"_465":1964,"_420":1965,"_450":1966},{"_450":1961},{"_1419":1944,"_1945":1946,"_1947":1948},{"_351":352},{"_568":1943},{"_426":568},{"_420":1957,"_426":965},"embeddings",{"_1010":1954},"files",{"_1010":1949},{"_465":1950,"_463":1951,"_1952":1953},"registry.astropods.ai/b552bc5f-4b64-47fa-8d33-2a3a0905d997/memory-box-knowledge-files:ebf00db6",9000,"volume","/data",{"_465":1955,"_463":1956},"mindthemath/nomic-embed-v1.5:slim",8080,[1958],{"_439":440,"_642":1959,"_61":1960},"memory_box","POSTGRES_DB",{"_452":1962},{"_463":1963},3001,"registry.astropods.ai/b552bc5f-4b64-47fa-8d33-2a3a0905d997/memory-box:ebf00db6",[1967],{"_452":24,"_454":242},{"_439":440,"_642":108,"_379":1968,"_61":1969},"Set to a new unique value and redeploy to clear the admin password (next login prompts setup).","PASSWORD_RESET_KEY","#95563d","#f7b9a1","#c14211","#86452d","#22120b","#e3ba9c","#e09e85","knowledge-search","https://assets.astropods.ai/avatars/agents/ankurkumar/knowledge-search.jpg?v=1787090497",{"_468":2007,"_470":2008,"_472":2009,"_474":475,"_476":2010,"_478":1974,"_480":875,"_482":2011,"_484":1976},[1984],{"_366":303,"_367":303},[1983],{"_61":180,"_347":43},{"_369":1985,"_371":1986,"_373":1987,"_375":1988,"_377":1989},"1721de73",{"_416":1999,"_418":2000,"_422":2001,"_371":424},"---\ndescription: \"Searches Vedcraft blog posts and answers questions from their full article content.\"\ntags: []\nauthors: []\ncapabilities:\n - \"Search Vedcraft posts by topic, title, taxonomy, author, or date\"\n - \"Read matching Vedcraft posts and cite their source URLs\"\nintegrations:\n - \"anthropic\"\n - \"vedcraft-mcp\"\n---\n\nKnowledge Search Astro Postman searches the public Vedcraft MCP server for relevant blog posts,\nreads matching articles when needed, and answers with Vedcraft source links.\n",{"_379":1990,"_385":1991,"_387":1992,"_389":1993},"2026-08-18T23:04:05Z","Searches Vedcraft blog posts and answers questions from their full article content.",[1997,1998],"\nKnowledge Search Astro Postman searches the public Vedcraft MCP server for relevant blog posts,\nreads matching articles when needed, and answers with Vedcraft source links.\n",[1994,1995],{"_59":568,"_61":569,"_397":24},{"_59":1996,"_61":1996,"_397":242},"vedcraft-mcp","Search Vedcraft posts by topic, title, taxonomy, author, or date","Read matching Vedcraft posts and cite their source URLs",{"_465":2006},{"_450":2003},{"_568":2002},{"_426":568},{"_454":2004},{"_456":2005},[462],"registry.astropods.ai/e0c5bea7-d8c6-465b-8cd6-0fb4acac99bf/knowledge-search:1721de73","#c88971","#f9b89f","#f47645","#86462d","#e6ae99","juuglmaps","https://assets.astropods.ai/avatars/agents/postman/juuglmaps.jpg?v=1790695228",{"_468":2072,"_470":2073,"_472":2074,"_474":475,"_476":2075,"_478":2076,"_480":710,"_482":2077,"_484":2078},[2021],{"_366":475,"_367":475},[2018],{"_61":2019,"_347":2020},"Max Friedrich","maxfriedrich",{"_369":2022,"_371":2023,"_373":2024,"_375":2025,"_377":2026},"e4cb2c90",{"_416":2052,"_418":2053,"_420":2054,"_422":2055,"_371":424},"---\ndescription: \"Answers questions about the JuuglMaps OpenStreetMap stack and calls its four public APIs to get the answer.\"\ntags:\n - openstreetmap\n - geospatial\n - api\n - read-only\nauthors:\n - name: Maximilian Friedrich\n account: maxfriedrich\ncapabilities:\n - Geocodes addresses and place names, and reverse-geocodes coordinates\n - Finds map features by tag, area or radius with Overpass QL\n - Computes driving routes, ETAs and distance matrices\n - Reads OSM nodes, their history and changesets\n - Answers documentation questions from the published Fern portal\nrepository: \"github:postman-solutions-eng/maxfrie-JuuglMaps\"\nintegrations:\n - JuuglMaps OpenStreetMap stack\n - Fern documentation MCP server\n---\n\n# JuuglMaps\n\nAn agent over the JuuglMaps OpenStreetMap stack: four public APIs — OSM Editing v0.6,\nNominatim geocoding, Overpass feature queries and OSRM routing — served from one origin\nover a Germany-only dataset.\n\nIt has two sets of tools. Eighteen tools call the APIs directly. One tool, `searchDocs`,\nsearches the published documentation portal for questions about the APIs themselves.\n\n## It cannot write to the map, by construction\n\nEvery OSM write — opening or closing a changeset, creating, updating or deleting a node —\nis absent and unreachable. This is not a prompt instruction the model could talk itself\nout of. The tool layer exposes only operations whose OpenAPI `security` is the empty\nlist, refuses any method other than `GET` (plus the single Overpass `POST`), and never\nattaches a credential header. A write needs a JuuglMaps gateway API key *and* an OSM\nOAuth token with `write_api` scope; this agent holds neither.\n\n## Why the API tools are hand-written\n\nFern's generated API Definition cannot express two things this stack depends on. The\nOverpass `text/plain` query body is dropped by the intermediate representation, so a\ngenerated tool would have no parameter to carry the query at all. And all eight XML\nresponses vanish from the Definition entirely. Both are recorded in the source\nrepository's own audit (`docs/OBSERVED.md` #36 and #42). The tools are therefore written\nagainst the OpenAPI documents directly, with the documented traps — coordinate order,\nbounding-box limits, the ambiguous 504 — encoded in the tool descriptions.\n\n## Usage\n\nAsk it map questions and it will make real requests:\n\n- \"What pharmacies are within 500 m of Marienplatz in Munich?\"\n- \"How long is the drive from Munich to Nuremberg?\"\n- \"What are the tags on node 1?\"\n- \"Which of these three addresses is closest by car to our office?\"\n\nAsk it API questions and it will search the portal:\n\n- \"How do I open a changeset?\"\n- \"What are the limits on a bounding-box download?\"\n\n## Limitations\n\n- **Germany only**, from a fixed extract. Outside Germany the honest answer is that this\n deployment has no data there.\n- **The origin serves Mon–Fri 07:00–19:00 Europe/Berlin.** Outside that window requests\n fail at the gateway; the agent reports that as an environment problem rather than as\n missing data.\n- **Vector tiles are not exposed.** `getRoutingTile` answers with binary protobuf\n (~170 KB), which cannot survive a text tool result.\n- **`searchDocs` refuses any query containing the literal word \"overpass\"** — a\n server-side input filter on the documentation service, reproduced deterministically and\n defeated by dropping one letter. The agent works around it by rephrasing; the\n `overpass_query` tool is unaffected.\n\n## Attribution\n\nData is © OpenStreetMap contributors, licensed under ODbL 1.0. Every API response carries\nan `X-Attribution` header. Downstream use must preserve attribution and applicable\nshare-alike obligations.\n",{"_379":2027,"_381":2028,"_559":2029,"_383":2030,"_385":2031,"_387":2032,"_389":2033},"2026-09-29T15:08:18Z","Answers questions about the JuuglMaps OpenStreetMap stack and calls its four public APIs to get the answer.",[2048,2049,2050,2051],[2046],{"_62":574,"_409":2045},[2040,2041,2042,2043,2044],"\n# JuuglMaps\n\nAn agent over the JuuglMaps OpenStreetMap stack: four public APIs — OSM Editing v0.6,\nNominatim geocoding, Overpass feature queries and OSRM routing — served from one origin\nover a Germany-only dataset.\n\nIt has two sets of tools. Eighteen tools call the APIs directly. One tool, `searchDocs`,\nsearches the published documentation portal for questions about the APIs themselves.\n\n## It cannot write to the map, by construction\n\nEvery OSM write — opening or closing a changeset, creating, updating or deleting a node —\nis absent and unreachable. This is not a prompt instruction the model could talk itself\nout of. The tool layer exposes only operations whose OpenAPI `security` is the empty\nlist, refuses any method other than `GET` (plus the single Overpass `POST`), and never\nattaches a credential header. A write needs a JuuglMaps gateway API key *and* an OSM\nOAuth token with `write_api` scope; this agent holds neither.\n\n## Why the API tools are hand-written\n\nFern's generated API Definition cannot express two things this stack depends on. The\nOverpass `text/plain` query body is dropped by the intermediate representation, so a\ngenerated tool would have no parameter to carry the query at all. And all eight XML\nresponses vanish from the Definition entirely. Both are recorded in the source\nrepository's own audit (`docs/OBSERVED.md` #36 and #42). The tools are therefore written\nagainst the OpenAPI documents directly, with the documented traps — coordinate order,\nbounding-box limits, the ambiguous 504 — encoded in the tool descriptions.\n\n## Usage\n\nAsk it map questions and it will make real requests:\n\n- \"What pharmacies are within 500 m of Marienplatz in Munich?\"\n- \"How long is the drive from Munich to Nuremberg?\"\n- \"What are the tags on node 1?\"\n- \"Which of these three addresses is closest by car to our office?\"\n\nAsk it API questions and it will search the portal:\n\n- \"How do I open a changeset?\"\n- \"What are the limits on a bounding-box download?\"\n\n## Limitations\n\n- **Germany only**, from a fixed extract. Outside Germany the honest answer is that this\n deployment has no data there.\n- **The origin serves Mon–Fri 07:00–19:00 Europe/Berlin.** Outside that window requests\n fail at the gateway; the agent reports that as an environment problem rather than as\n missing data.\n- **Vector tiles are not exposed.** `getRoutingTile` answers with binary protobuf\n (~170 KB), which cannot survive a text tool result.\n- **`searchDocs` refuses any query containing the literal word \"overpass\"** — a\n server-side input filter on the documentation service, reproduced deterministically and\n defeated by dropping one letter. The agent works around it by rephrasing; the\n `overpass_query` tool is unaffected.\n\n## Attribution\n\nData is © OpenStreetMap contributors, licensed under ODbL 1.0. Every API response carries\nan `X-Attribution` header. Downstream use must preserve attribution and applicable\nshare-alike obligations.\n",[2034,2035],{"_59":2038,"_61":2039,"_397":242},{"_59":2036,"_61":2037,"_397":242},"fern documentation mcp server","Fern documentation MCP server","juuglmaps openstreetmap stack","JuuglMaps OpenStreetMap stack","Geocodes addresses and place names, and reverse-geocodes coordinates","Finds map features by tag, area or radius with Overpass QL","Computes driving routes, ETAs and distance matrices","Reads OSM nodes, their history and changesets","Answers documentation questions from the published Fern portal","https://github.com/postman-solutions-eng/maxfrie-JuuglMaps",{"_61":2047,"_347":2020},"Maximilian Friedrich","openstreetmap","geospatial","api","read-only",{"_465":2071},{"_450":2068},{"_2058":2059,"_2060":2061},{"_642":2056},{"_422":2057,"_426":645},[647],"base-url",{"_439":440,"_642":2065,"_379":2066,"_442":443,"_61":2067,"_984":24,"_445":242},"docs-mcp-url",{"_439":440,"_642":2062,"_379":2063,"_442":443,"_61":2064,"_984":24,"_445":242},"https://fern-osm-demo.docs.buildwithfern.com/_mcp/server","Fern-hosted documentation MCP endpoint, which supplies the searchDocs tool. If it is unreachable the agent still starts with the eighteen API tools.","JUUGLMAPS_DOCS_MCP_URL","https://dgm5tvjbc32h9.cloudfront.net","Origin for the four APIs. Reads need no credential, and writes are not reachable from this agent.","JUUGLMAPS_BASE_URL",{"_454":2069},{"_456":2070},[462],"registry.astropods.ai/postman/juuglmaps:e4cb2c90","#3e7b8b","#a5e3f3","#1792b1","#2d7486","#0b1e22","#99dee6","#85cee0","industry-news-agent","https://assets.astropods.ai/avatars/agents/astro-ai/industry-news-agent.jpg",{"_468":2143,"_470":2144,"_472":2145,"_474":475,"_476":2146,"_478":2147,"_480":875,"_482":2148,"_484":2149},[2087],{"_366":112,"_367":2086},[2085],{"_61":192,"_347":41},6,{"_369":2088,"_371":2089,"_373":2090,"_375":2091,"_377":2092},"5bda2a8a",{"_416":2120,"_418":2121,"_420":2122,"_422":2123,"_371":424},"---\ndescription: \"Fetches industry news from configured sources in parallel, deduplicates, and delivers an AI briefing via web or Slack.\"\ntags:\n - news\n - industry-intelligence\n - openai\n - product-management\n - research\ncapabilities:\n - \"Fetch news in parallel from any configured sources (NewsAPI, GNews, The Guardian, MediaStack)\"\n - \"Deduplicate articles across sources by title\"\n - \"Summarise all deduplicated articles into a structured briefing using GPT-4o mini\"\n - \"Deliver output as summary, deep analysis, or key insights\"\n - \"Work with any combination of API keys — all sources are optional\"\n - \"Report clearly when no news sources are configured\"\nrepository:\n type: github\n url: https://github.com/astropods/agents\n directory: industry-news-agent\nintegrations:\n - OpenAI\n - NewsAPI\n - GNews\n - The Guardian\n - MediaStack\n---\n\n# Industry News Agent\n\nKeeping up with your industry means checking four different news sites every morning. Industry News Agent does it in one shot — fetching from NewsAPI, GNews, The Guardian, and MediaStack in parallel, removing duplicates, and handing the result to GPT-4o mini for a structured briefing. Ask for a summary, a deep analysis, or just the key insights.\n\n## Usage\n\nSend a topic — optionally append a format keyword:\n\n| Message | Effect |\n|---------|--------|\n| `AI news` | Key themes, top stories, takeaway (default summary) |\n| `startup funding analysis` | Market signals, key players, risks & opportunities |\n| `fintech key insights` | 5-7 actionable bullet points + what to watch |\n\n**Examples:**\n- *\"electric vehicles Europe\"* — latest EV news summarised\n- *\"quantum computing analysis\"* — deep analytical breakdown\n- *\"SaaS pricing key insights\"* — direct, actionable bullets\n\n## Output formats\n\n| Format | Triggered by | Structure |\n|--------|-------------|-----------|\n| **Summary** | default | Key themes · Top stories · Takeaway |\n| **Analysis** | `analysis` / `analyse` / `analyze` | Market signals · Key players · Risks & opportunities · Verdict |\n| **Key insights** | `key insights` / `insights` | 5-7 action-verb bullets · What to watch |\n\n## Sources\n\n| Source | Key required | Get a key |\n|--------|-------------|-----------|\n| NewsAPI | `NEWS_API_KEY` | [newsapi.org](https://newsapi.org) |\n| GNews | `GNEWS_API_KEY` | [gnews.io](https://gnews.io) |\n| The Guardian | `GUARDIAN_API_KEY` | [open-platform.theguardian.com](https://open-platform.theguardian.com/access/) |\n| MediaStack | `MEDIASTACK_API_KEY` | [mediastack.com](https://mediastack.com) |\n\nAll four API keys are **optional** — configure any combination and the agent will only query the sources you've set up. If none are configured, the agent will tell you which keys to add. Sources that are configured but encounter an error (quota exceeded, outage) are skipped gracefully.\n\n## Environment variables\n\n| Variable | Required | Description |\n|----------|----------|-------------|\n| `NEWS_API_KEY` | Optional | NewsAPI key — configured at deploy time |\n| `GNEWS_API_KEY` | Optional | GNews API key — configured at deploy time |\n| `GUARDIAN_API_KEY` | Optional | The Guardian Open Platform key — configured at deploy time |\n| `MEDIASTACK_API_KEY` | Optional | MediaStack API key — configured at deploy time |\n| `OPENAI_API_KEY` | Required | Auto-injected by Astropods |\n\n## Slack integration\n\nThe agent supports both the **web** and **Slack** adapters. At deploy time, choose which adapter(s) to enable — the Slack adapter handles authentication and delivery automatically.\n\n## Limitations\n\n- Up to 10 articles per source (40 total before dedup); all deduplicated articles are sent to the summariser.\n- Article descriptions are truncated at 200 characters before being sent to OpenAI.\n- Deduplication matches on exact (lowercased, trimmed) title only — near-duplicate headlines are not merged.\n- The full briefing is returned once all sources are queried and summarised; per-source progress is not streamed.\n",{"_379":2093,"_381":2094,"_383":2095,"_385":2096,"_387":2097,"_389":2098},"2026-09-28T19:32:55Z","Fetches industry news from configured sources in parallel, deduplicates, and delivers an AI briefing via web or Slack.",[2118,2119,402,519,1621],{"_62":408,"_409":410,"_411":2079},[2112,2113,2114,2115,2116,2117],"\n# Industry News Agent\n\nKeeping up with your industry means checking four different news sites every morning. Industry News Agent does it in one shot — fetching from NewsAPI, GNews, The Guardian, and MediaStack in parallel, removing duplicates, and handing the result to GPT-4o mini for a structured briefing. Ask for a summary, a deep analysis, or just the key insights.\n\n## Usage\n\nSend a topic — optionally append a format keyword:\n\n| Message | Effect |\n|---------|--------|\n| `AI news` | Key themes, top stories, takeaway (default summary) |\n| `startup funding analysis` | Market signals, key players, risks & opportunities |\n| `fintech key insights` | 5-7 actionable bullet points + what to watch |\n\n**Examples:**\n- *\"electric vehicles Europe\"* — latest EV news summarised\n- *\"quantum computing analysis\"* — deep analytical breakdown\n- *\"SaaS pricing key insights\"* — direct, actionable bullets\n\n## Output formats\n\n| Format | Triggered by | Structure |\n|--------|-------------|-----------|\n| **Summary** | default | Key themes · Top stories · Takeaway |\n| **Analysis** | `analysis` / `analyse` / `analyze` | Market signals · Key players · Risks & opportunities · Verdict |\n| **Key insights** | `key insights` / `insights` | 5-7 action-verb bullets · What to watch |\n\n## Sources\n\n| Source | Key required | Get a key |\n|--------|-------------|-----------|\n| NewsAPI | `NEWS_API_KEY` | [newsapi.org](https://newsapi.org) |\n| GNews | `GNEWS_API_KEY` | [gnews.io](https://gnews.io) |\n| The Guardian | `GUARDIAN_API_KEY` | [open-platform.theguardian.com](https://open-platform.theguardian.com/access/) |\n| MediaStack | `MEDIASTACK_API_KEY` | [mediastack.com](https://mediastack.com) |\n\nAll four API keys are **optional** — configure any combination and the agent will only query the sources you've set up. If none are configured, the agent will tell you which keys to add. Sources that are configured but encounter an error (quota exceeded, outage) are skipped gracefully.\n\n## Environment variables\n\n| Variable | Required | Description |\n|----------|----------|-------------|\n| `NEWS_API_KEY` | Optional | NewsAPI key — configured at deploy time |\n| `GNEWS_API_KEY` | Optional | GNews API key — configured at deploy time |\n| `GUARDIAN_API_KEY` | Optional | The Guardian Open Platform key — configured at deploy time |\n| `MEDIASTACK_API_KEY` | Optional | MediaStack API key — configured at deploy time |\n| `OPENAI_API_KEY` | Required | Auto-injected by Astropods |\n\n## Slack integration\n\nThe agent supports both the **web** and **Slack** adapters. At deploy time, choose which adapter(s) to enable — the Slack adapter handles authentication and delivery automatically.\n\n## Limitations\n\n- Up to 10 articles per source (40 total before dedup); all deduplicated articles are sent to the summariser.\n- Article descriptions are truncated at 200 characters before being sent to OpenAI.\n- Deduplication matches on exact (lowercased, trimmed) title only — near-duplicate headlines are not merged.\n- The full briefing is returned once all sources are queried and summarised; per-source progress is not streamed.\n",[2099,2100,2101,2102,2103],{"_59":402,"_61":403,"_397":24},{"_59":2110,"_61":2111,"_397":242},{"_59":2108,"_61":2109,"_397":242},{"_59":2106,"_61":2107,"_397":242},{"_59":2104,"_61":2105,"_397":242},"mediastack","MediaStack","the guardian","The Guardian","gnews","GNews","newsapi","NewsAPI","Fetch news in parallel from any configured sources (NewsAPI, GNews, The Guardian, MediaStack)","Deduplicate articles across sources by title","Summarise all deduplicated articles into a structured briefing using GPT-4o mini","Deliver output as summary, deep analysis, or key insights","Work with any combination of API keys — all sources are optional","Report clearly when no news sources are configured","news","industry-intelligence",{"_465":2141,"_450":2142},{"_450":2137},{"_2125":2126,"_2127":2128,"_2129":2130,"_2131":2132},{"_402":2124},{"_426":402},"GNEWS_API_KEY",{"_439":440,"_379":2136,"_442":443,"_61":2125,"_984":24,"_445":24},"GUARDIAN_API_KEY",{"_439":440,"_379":2135,"_442":443,"_61":2127,"_984":24,"_445":24},"MEDIASTACK_API_KEY",{"_439":440,"_379":2134,"_442":443,"_61":2129,"_984":24,"_445":24},"NEWS_API_KEY",{"_439":440,"_379":2133,"_442":443,"_61":2131,"_984":24,"_445":24},"NewsAPI key (newsapi.org)","MediaStack API key (mediastack.com)","The Guardian Open Platform API key (open-platform.theguardian.com)","GNews API key (gnews.io)",{"_454":2138},{"_456":2139,"_458":2140},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/industry-news-agent:5bda2a8a",{"_454":24},"#c2bab9","#d3c7c5","#c6b7b4","#65514d","#1a1514","#c8b9b7","#bdaca8","incident-manager","https://assets.astropods.ai/avatars/agents/astro-ai/incident-manager.jpg",{"_468":2196,"_470":2197,"_472":2198,"_474":475,"_476":2199,"_478":874,"_480":875,"_482":2200,"_484":2201},[2157],{"_366":303,"_367":303},[2156],{"_61":192,"_347":41},{"_369":2158,"_371":2159,"_373":2160,"_375":2161,"_377":2162},"0320c9b4",{"_416":2179,"_418":2180,"_420":2181,"_422":2182,"_371":424},"---\ndescription: \"Slack-triggered incident management agent that tracks incidents in Notion and keeps summaries updated in real time.\"\ntags:\n - slack\n - notion\n - incident-management\n - openai\n - devops\ncapabilities:\n - \"Create a new incident record in Notion when triggered by an @mention in Slack\"\n - \"Track all incidents in a Notion database with status and severity\"\n - \"Generate and update detail summary, engineering update, and support update per incident\"\n - \"Detect duplicate incidents based on timestamps and existing Notion records\"\nrepository:\n type: github\n url: https://github.com/astropods/agents\n directory: incident-management-agent\nintegrations:\n - OpenAI\n - Slack (via built-in adapter)\n - Notion\n---\n\n# Incident Manager Agent\n\nAn incident is declared in Slack — the agent takes it from there. It logs the incident in Notion and keeps the detail summary, engineering update, and support update current as the conversation unfolds.\n\n## Triggers\n\n- **@mention in Slack** (`@incident-manager start incident: `): Creates a new incident page in Notion.\n- **Follow-up @mentions**: Uses conversation context to determine whether to create a new incident or update an existing one.\n\n## Tools\n\n| Tool | Description |\n|---|---|\n| `fetch_list_of_incidents_from_notion` | Lists all incidents from the Notion database |\n| `create_new_incident_in_notion` | Creates a new incident page in Notion |\n| `update_notion_incident_page` | Updates status, severity, and summary blocks on an existing page |\n\n## Environment variables\n\n| Variable | Description |\n|---|---|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `NOTION_API_KEY` | Notion integration secret |\n| `NOTION_DATABASE_ID` | Notion database ID for the incident log (treated as a secret) |\n",{"_379":2163,"_381":2164,"_383":2165,"_385":2166,"_387":2167,"_389":2168},"2026-09-28T19:35:29Z","Slack-triggered incident management agent that tracks incidents in Notion and keeps summaries updated in real time.",[458,800,2177,402,2178],{"_62":408,"_409":410,"_411":2176},[2172,2173,2174,2175],"\n# Incident Manager Agent\n\nAn incident is declared in Slack — the agent takes it from there. It logs the incident in Notion and keeps the detail summary, engineering update, and support update current as the conversation unfolds.\n\n## Triggers\n\n- **@mention in Slack** (`@incident-manager start incident: `): Creates a new incident page in Notion.\n- **Follow-up @mentions**: Uses conversation context to determine whether to create a new incident or update an existing one.\n\n## Tools\n\n| Tool | Description |\n|---|---|\n| `fetch_list_of_incidents_from_notion` | Lists all incidents from the Notion database |\n| `create_new_incident_in_notion` | Creates a new incident page in Notion |\n| `update_notion_incident_page` | Updates status, severity, and summary blocks on an existing page |\n\n## Environment variables\n\n| Variable | Description |\n|---|---|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `NOTION_API_KEY` | Notion integration secret |\n| `NOTION_DATABASE_ID` | Notion database ID for the incident log (treated as a secret) |\n",[2169,2170,2171],{"_59":402,"_61":403,"_397":24},{"_59":395,"_61":396,"_397":242},{"_59":800,"_61":801,"_397":24},"Create a new incident record in Notion when triggered by an @mention in Slack","Track all incidents in a Notion database with status and severity","Generate and update detail summary, engineering update, and support update per incident","Detect duplicate incidents based on timestamps and existing Notion records","incident-management-agent","incident-management","devops",{"_465":2194,"_450":2195},{"_450":2190},{"_2184":2185,"_2186":2187},{"_402":2183},{"_426":402},"NOTION_API_KEY",{"_439":440,"_379":2189,"_442":443,"_61":2184,"_445":24},"NOTION_DATABASE_ID",{"_439":440,"_379":2188,"_442":443,"_61":2186,"_445":24},"Notion database ID holding the incident log","Notion integration secret",{"_454":2191},{"_456":2192,"_458":2193},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/incident-manager:0320c9b4",{"_454":24},"#763b40","#eeaaaf","#941d27","#862d34","#d9a6aa","#e0858c","github-oauth-agent","registry.astropods.ai/pmrr","https://assets.astropods.ai/avatars/agents/rrabbah/github-oauth-agent.jpg",{"_468":2274,"_470":2275,"_472":2276,"_474":475,"_476":2277,"_478":2278,"_480":2279,"_482":2280,"_484":2281},[2210],{"_366":1721,"_367":493},[2209],{"_61":192,"_347":41},{"_369":2211,"_371":2212,"_373":2213,"_375":2214,"_377":2215},"39e80787",{"_416":2237,"_418":2238,"_862":2239,"_422":2240,"_371":584},"---\ndescription: \"OAuth2-authenticated AI agent that gates every conversation behind GitHub sign-in and answers questions about the user's repos\"\ntags:\n - oauth2\n - github\n - openai\n - mastra\n - example\nauthors:\n - name: rabbah\n account: rabbah\ncapabilities:\n - \"Enforces OAuth2 authentication before processing any conversation message\"\n - \"Exchanges OAuth authorization codes and manages sessions with TTL-based expiry\"\n - \"Injects authenticated user identity (ID, name, email) into conversation context\"\n - \"Lists the authenticated user's GitHub repositories\"\n - \"Retrieves repository metadata (stars, forks, open issues, language, topics)\"\n - \"Fetches and reads README files from any of the user's repositories\"\n - \"Displays recent GitHub activity (pushes, PRs, issues, forks) for the user\"\n - \"Supports OIDC and GitHub OAuth provider field conventions\"\nrepository: github:rabbah/github-oauth-agent\nintegrations:\n - GitHub\n - OpenAI\n - Mastra\n---\n\n## Overview\n\n`github-oauth-agent` demonstrates a complete OAuth2 authentication flow for AI agents built on the Astropods platform. It wraps any underlying agent adapter with an `AuthAdapter` that intercepts every incoming message: unauthenticated users receive a sign-in link, while authenticated users have their identity automatically injected into the conversation prompt.\n\nSessions are stored in-memory with an 8-hour TTL and keyed by conversation ID, allowing the agent to maintain per-conversation auth state across turns.\n\nThe agent uses GitHub as the Identity Provider and operates on behalf of the authenticated user to list repositories, fetch README content, retrieve repo metadata, and surface recent activity — making it a useful reference implementation for any agent that needs to act as the signed-in user.\n\n## Usage\n\n1. Configure OAuth credentials and endpoints via the required inputs (`OAUTH_CLIENT_ID`, `OAUTH_CLIENT_SECRET`, `OAUTH_AUTH_URL`, `OAUTH_TOKEN_URL`, `OAUTH_CALLBACK_URL`).\n2. On first message, the agent returns a sign-in link. Click it to authorize via GitHub.\n3. After authorization, GitHub redirects to the callback server, which exchanges the code for a token, fetches user info, and creates a session.\n4. Subsequent messages are processed normally with the user's identity prepended to the prompt.\n\n### Required Inputs\n\n| Input | Description |\n|---|---|\n| `OAUTH_CLIENT_ID` | OAuth client ID |\n| `OAUTH_CLIENT_SECRET` | OAuth client secret |\n| `OAUTH_AUTH_URL` | Authorization endpoint URL |\n| `OAUTH_TOKEN_URL` | Token exchange endpoint URL |\n| `OAUTH_CALLBACK_URL` | Redirect URI registered with your OAuth provider |\n\n### Optional Inputs\n\n| Input | Default | Description |\n|---|---|---|\n| `OAUTH_SCOPES` | `openid profile email` | Scopes to request |\n| `OAUTH_USERINFO_URL` | — | Userinfo endpoint (OIDC); falls back to token sub-claim |\n\n### Example Prompts\n\nOnce authenticated, you can ask things like:\n\n- *\"List my repositories\"*\n- *\"Show me my recent GitHub activity\"*\n- *\"How many open issues does my repo have?\"*\n\n## Limitations\n\n- Sessions are stored in-memory; restarting the agent clears all active sessions and requires users to re-authenticate.\n- The agent uses a single OAuth client configuration — it does not support multi-tenant or per-user OAuth app credentials.\n- Token refresh is not implemented; sessions expire after 8 hours and the user must sign in again.\n",{"_379":2216,"_381":2217,"_559":2218,"_383":2219,"_385":2220,"_387":2221,"_389":2222},"2026-05-13T15:33:23Z","OAuth2-authenticated AI agent that gates every conversation behind GitHub sign-in and answers questions about the user's repos",[2236,408,402,579,580],[2235],{"_62":574,"_409":2234},[2226,2227,2228,2229,2230,2231,2232,2233],"\n## Overview\n\n`github-oauth-agent` demonstrates a complete OAuth2 authentication flow for AI agents built on the Astropods platform. It wraps any underlying agent adapter with an `AuthAdapter` that intercepts every incoming message: unauthenticated users receive a sign-in link, while authenticated users have their identity automatically injected into the conversation prompt.\n\nSessions are stored in-memory with an 8-hour TTL and keyed by conversation ID, allowing the agent to maintain per-conversation auth state across turns.\n\nThe agent uses GitHub as the Identity Provider and operates on behalf of the authenticated user to list repositories, fetch README content, retrieve repo metadata, and surface recent activity — making it a useful reference implementation for any agent that needs to act as the signed-in user.\n\n## Usage\n\n1. Configure OAuth credentials and endpoints via the required inputs (`OAUTH_CLIENT_ID`, `OAUTH_CLIENT_SECRET`, `OAUTH_AUTH_URL`, `OAUTH_TOKEN_URL`, `OAUTH_CALLBACK_URL`).\n2. On first message, the agent returns a sign-in link. Click it to authorize via GitHub.\n3. After authorization, GitHub redirects to the callback server, which exchanges the code for a token, fetches user info, and creates a session.\n4. Subsequent messages are processed normally with the user's identity prepended to the prompt.\n\n### Required Inputs\n\n| Input | Description |\n|---|---|\n| `OAUTH_CLIENT_ID` | OAuth client ID |\n| `OAUTH_CLIENT_SECRET` | OAuth client secret |\n| `OAUTH_AUTH_URL` | Authorization endpoint URL |\n| `OAUTH_TOKEN_URL` | Token exchange endpoint URL |\n| `OAUTH_CALLBACK_URL` | Redirect URI registered with your OAuth provider |\n\n### Optional Inputs\n\n| Input | Default | Description |\n|---|---|---|\n| `OAUTH_SCOPES` | `openid profile email` | Scopes to request |\n| `OAUTH_USERINFO_URL` | — | Userinfo endpoint (OIDC); falls back to token sub-claim |\n\n### Example Prompts\n\nOnce authenticated, you can ask things like:\n\n- *\"List my repositories\"*\n- *\"Show me my recent GitHub activity\"*\n- *\"How many open issues does my repo have?\"*\n\n## Limitations\n\n- Sessions are stored in-memory; restarting the agent clears all active sessions and requires users to re-authenticate.\n- The agent uses a single OAuth client configuration — it does not support multi-tenant or per-user OAuth app credentials.\n- Token refresh is not implemented; sessions expire after 8 hours and the user must sign in again.\n",[2223,2224,2225],{"_59":408,"_61":1337,"_397":24},{"_59":402,"_61":403,"_397":24},{"_59":579,"_61":1082,"_397":242},"Enforces OAuth2 authentication before processing any conversation message","Exchanges OAuth authorization codes and manages sessions with TTL-based expiry","Injects authenticated user identity (ID, name, email) into conversation context","Lists the authenticated user's GitHub repositories","Retrieves repository metadata (stars, forks, open issues, language, topics)","Fetches and reads README files from any of the user's repositories","Displays recent GitHub activity (pushes, PRs, issues, forks) for the user","Supports OIDC and GitHub OAuth provider field conventions","https://github.com/rabbah/github-oauth-agent",{"_61":37,"_347":37},"oauth2",{"_465":2246,"_420":2247,"_450":2248},{"_586":587,"_450":2242},{"_351":352},{"_402":2241},{"_426":402},{"_452":2243,"_454":2244},{"_463":1113},{"_456":2245},[462],"registry.astropods.ai/pmrr/github-oauth-agent:39e80787",[2249,2250,2251,2252,2253,2254,2255],{"_452":24,"_454":24},{"_439":440,"_379":2272,"_61":2273},{"_439":440,"_379":2270,"_61":2271,"_445":24},{"_439":440,"_642":2267,"_379":2268,"_61":2269},{"_439":440,"_642":2264,"_379":2265,"_61":2266},{"_439":440,"_379":2262,"_61":2263},{"_439":440,"_642":2259,"_379":2260,"_61":2261,"_984":24},{"_439":440,"_642":2256,"_379":2257,"_61":2258,"_984":24},"https://api.github.com/user","OAuth userinfo endpoint URL","OAUTH_USERINFO_URL","openid profile email","OAuth scopes to request","OAUTH_SCOPES","OAuth redirect URI","OAUTH_CALLBACK_URL","https://github.com/login/oauth/access_token","OAuth token exchange endpoint URL","OAUTH_TOKEN_URL","https://github.com/login/oauth/authorize","OAuth authorization endpoint URL","OAUTH_AUTH_URL","OAuth client secret","OAUTH_CLIENT_SECRET","OAuth client ID","OAUTH_CLIENT_ID","#234f4d","#a6f2f0","#0e6461","#2d8683","#0b2222","#f4f6f6","#e699ca","#85e0dd","github-issue-scorer-jev","registry.astropods.ai/astro-corp","https://assets.astropods.ai/avatars/agents/astro-corp/github-issue-scorer-jev.jpg?v=1790169090",{"_468":2346,"_470":2347,"_472":2348,"_474":475,"_476":2349,"_478":2350,"_480":481,"_482":2351,"_484":2352},[2290],{"_366":303,"_367":493},[2289],{"_61":31,"_347":31},{"_369":2291,"_371":2292,"_373":2293,"_375":2294,"_377":2295},"9a425ea9",{"_416":2325,"_418":2326,"_389":2327,"_862":2328,"_371":584},"---\ndescription: \"Triages a repo's open GitHub issues by priority and sentiment, streamed live into a dashboard — with a calibrated confidence on every score, from Jev.\"\ntags:\n - github\n - triage\n - scoring\n - product-management\n - sentiment-analysis\n - dashboard\n - jev\n - ag-ui\n - adk\nauthors:\n - name: Simon Guerrier\n account: simon\nrepository:\n type: git\n url: https://github.com/Simwar/astro-demo.git\n directory: github-issue-scorer-adk\ncapabilities:\n - \"Triage a repo's open GitHub issues by priority, sentiment, and user impact\"\n - \"Show how confident the model is in each score, and flag the shaky ones\"\n - \"Rank issues by severity so the top of the list is the right place to start\"\n - \"Surface competitor mentions and workarounds buried in comment threads\"\n - \"Fill in a live board as each issue is scored, rather than waiting for a report\"\n - \"Score one issue by number, or up to 50 at a time\"\nintegrations:\n - GitHub\n - OpenAI\n - TypeSafe Jev\n - Postman Fabric Gateway\n---\n\n# GitHub Issue Scorer\n\nYour backlog has four hundred open issues and you have Tuesday afternoon. Which\nthree actually matter?\n\nPoint this agent at a repository and it reads each open issue — the body and the\nwhole comment thread — then fills a live board as it goes. Every card tells you\nhow urgent the issue is, how the reporter feels about it, and how badly users are\nactually impacted. The frustrated thread where someone quietly posted a\nworkaround, or mentioned they're evaluating a competitor, stops being invisible.\n\n## What you get\n\n- **A board that fills in as it thinks** — cards appear immediately as\n placeholders and resolve one by one, so you can start reading the first result\n while the rest are still being scored.\n- **Scores you can calibrate your trust against** — every priority and sentiment\n carries a confidence figure. Anything under 60% is highlighted, so a shaky\n judgement invites a second look instead of sitting there looking authoritative.\n Hover it for the full breakdown.\n- **A real ranking, not three buckets** — an impact score orders issues *within*\n each priority, so the top of a `HIGH` run is genuinely where to start. Each\n card says what its impact level means in plain words.\n- **The things a summary usually loses** — competitor comparisons and\n user-discovered workarounds get pulled out of the thread. And when the agent\n detects one it couldn't cleanly quote, the card says so rather than showing\n nothing — a nudge to go read that thread yourself.\n\n## Usage\n\nOpen the dashboard, type a repository, choose how many issues to score, and hit\nthe button.\n\n| You want | Do this |\n|---|---|\n| The 5 most recent open issues | Enter `pallets/flask`, leave the count at 5 |\n| A deeper sweep | Enter `vercel/next.js`, set the count to 20 |\n| One specific issue | Enter `rails/rails#50234` |\n\nCards appear as placeholders the moment a run starts, then resolve one at a time.\n\nCards sort themselves highest-priority first, then by impact within each\npriority, so the list arrives in the order you should work through it.\n\n## How the scoring works\n\nTwo models, each doing the half it is actually good at.\n\n**Jev** makes the judgements — priority, sentiment, and impact. It is a decision\nmodel rather than a chat model: it picks from a fixed set of answers and returns\na probability for each, so it cannot invent a priority that doesn't exist or\nmangle its own output. That is where the confidence figures come from.\n\n**An LLM** does the writing — the summary, the reasoning behind each score, and\npulling competitor names and workarounds out of the comments.\n\nThey run at the same time, so you don't wait for both. If Jev is unavailable the\nagent keeps working and falls back to the LLM's own judgement; you'll just see\nthe confidence and impact figures disappear from the cards.\n\n## Limitations\n\n- **Read-only.** It analyses issues; it never comments, labels, or closes\n anything on GitHub.\n- **Open issues only.** Closed issues and pull requests are skipped.\n- **Up to 50 issues per run**, and it only sees what the connected GitHub\n account can see — private repos need access.\n- **Very long threads get trimmed.** Issue bodies and individual comments are\n truncated, and only the first 120 comments are scored, so a thousand-comment\n epic will lose some nuance.\n- **Sessions don't persist.** Scores live as long as the page is open; reloading\n starts fresh.\n",{"_379":2296,"_381":2297,"_559":2298,"_383":2299,"_385":2300,"_387":2301,"_389":2302},"2026-09-23T13:11:30Z","Triages a repo's open GitHub issues by priority and sentiment, streamed live into a dashboard — with a calibrated confidence on every score, from Jev.",[408,414,2320,519,516,2321,2322,2323,2324],[2319],{"_62":574,"_409":2317,"_411":2318},[2311,2312,2313,2314,2315,2316],"\n# GitHub Issue Scorer\n\nYour backlog has four hundred open issues and you have Tuesday afternoon. Which\nthree actually matter?\n\nPoint this agent at a repository and it reads each open issue — the body and the\nwhole comment thread — then fills a live board as it goes. Every card tells you\nhow urgent the issue is, how the reporter feels about it, and how badly users are\nactually impacted. The frustrated thread where someone quietly posted a\nworkaround, or mentioned they're evaluating a competitor, stops being invisible.\n\n## What you get\n\n- **A board that fills in as it thinks** — cards appear immediately as\n placeholders and resolve one by one, so you can start reading the first result\n while the rest are still being scored.\n- **Scores you can calibrate your trust against** — every priority and sentiment\n carries a confidence figure. Anything under 60% is highlighted, so a shaky\n judgement invites a second look instead of sitting there looking authoritative.\n Hover it for the full breakdown.\n- **A real ranking, not three buckets** — an impact score orders issues *within*\n each priority, so the top of a `HIGH` run is genuinely where to start. Each\n card says what its impact level means in plain words.\n- **The things a summary usually loses** — competitor comparisons and\n user-discovered workarounds get pulled out of the thread. And when the agent\n detects one it couldn't cleanly quote, the card says so rather than showing\n nothing — a nudge to go read that thread yourself.\n\n## Usage\n\nOpen the dashboard, type a repository, choose how many issues to score, and hit\nthe button.\n\n| You want | Do this |\n|---|---|\n| The 5 most recent open issues | Enter `pallets/flask`, leave the count at 5 |\n| A deeper sweep | Enter `vercel/next.js`, set the count to 20 |\n| One specific issue | Enter `rails/rails#50234` |\n\nCards appear as placeholders the moment a run starts, then resolve one at a time.\n\nCards sort themselves highest-priority first, then by impact within each\npriority, so the list arrives in the order you should work through it.\n\n## How the scoring works\n\nTwo models, each doing the half it is actually good at.\n\n**Jev** makes the judgements — priority, sentiment, and impact. It is a decision\nmodel rather than a chat model: it picks from a fixed set of answers and returns\na probability for each, so it cannot invent a priority that doesn't exist or\nmangle its own output. That is where the confidence figures come from.\n\n**An LLM** does the writing — the summary, the reasoning behind each score, and\npulling competitor names and workarounds out of the comments.\n\nThey run at the same time, so you don't wait for both. If Jev is unavailable the\nagent keeps working and falls back to the LLM's own judgement; you'll just see\nthe confidence and impact figures disappear from the cards.\n\n## Limitations\n\n- **Read-only.** It analyses issues; it never comments, labels, or closes\n anything on GitHub.\n- **Open issues only.** Closed issues and pull requests are skipped.\n- **Up to 50 issues per run**, and it only sees what the connected GitHub\n account can see — private repos need access.\n- **Very long threads get trimmed.** Issue bodies and individual comments are\n truncated, and only the first 120 comments are scored, so a thousand-comment\n epic will lose some nuance.\n- **Sessions don't persist.** Scores live as long as the page is open; reloading\n starts fresh.\n",[2303,2304,2305,2306],{"_59":408,"_61":1337,"_397":24},{"_59":402,"_61":403,"_397":24},{"_59":2309,"_61":2310,"_397":242},{"_59":2307,"_61":2308,"_397":242},"postman fabric gateway","Postman Fabric Gateway","typesafe jev","TypeSafe Jev","Triage a repo's open GitHub issues by priority, sentiment, and user impact","Show how confident the model is in each score, and flag the shaky ones","Rank issues by severity so the top of the list is the right place to start","Surface competitor mentions and workarounds buried in comment threads","Fill in a live board as each issue is scored, rather than waiting for a report","Score one issue by number, or up to 50 at a time","https://github.com/Simwar/astro-demo.git","github-issue-scorer-adk",{"_61":908,"_347":31},"scoring","dashboard","jev","ag-ui","adk",{"_465":2333,"_420":2334,"_450":2335},{"_450":2330},{"_408":2329},{},{"_426":408},{"_452":2331},{"_463":2332},8808,"registry.astropods.ai/astro-corp/github-issue-scorer-jev:9a425ea9",[2336,2337,2338],{"_452":24},{"_439":440,"_379":2344,"_61":2345,"_445":24},{"_439":440,"_379":2342,"_61":2343},{"_439":440,"_642":2339,"_379":2340,"_61":2341},"1","Set to 0 to score with the LLM only, for comparison against Jev.","JEV_ENABLED","Fabric Gateway base URL.","FABRIC_GATEWAY_URL","Postman Fabric Gateway key, used for both the openai and jev routes.","FABRIC_GATEWAY_KEY","#3e4767","#b2bde6","#293b7c","#2d4086","#0b1022","#d9aaa5","#8c9dd9","github-issue-scorer","https://assets.astropods.ai/avatars/agents/astro-ai/github-issue-scorer.jpg",{"_468":2393,"_470":2394,"_472":2395,"_474":475,"_476":2396,"_478":538,"_480":539,"_482":2397,"_484":2398},[2360],{"_366":493,"_367":493},[2359],{"_61":192,"_347":41},{"_369":2361,"_371":2362,"_373":2363,"_375":2364,"_377":2365},"59cc69cd",{"_416":2381,"_418":2382,"_389":2383,"_422":2384,"_371":424},"---\ndescription: \"Scores GitHub issues by priority and sentiment using GPT-4o mini, surfacing competitor mentions and workarounds — via web or Slack.\"\ntags:\n - github\n - triage\n - sentiment-analysis\n - slack\n - product-management\n - openai\n - issues\ncapabilities:\n - \"Fetch and analyse up to 50 open GitHub issues in a single run\"\n - \"Score each issue HIGH, MEDIUM, or LOW priority using GPT-4o mini\"\n - \"Detect user sentiment: frustration, urgency, neutral, or positive\"\n - \"Surface competitor mentions and user-reported workarounds\"\n - \"Deliver ranked priority report via web chat or Slack adapter\"\n - \"Analyse a single issue by number (owner/repo#123)\"\nrepository:\n type: github\n url: https://github.com/astropods/agents\n directory: github-issue-scorer\nintegrations:\n - GitHub\n - OpenAI\n---\n\n# GitHub Issue Scorer\n\nYour backlog is full of open issues and you don't know where to start. GitHub Issue Scorer reads every issue — body and all comments — then uses GPT-4o mini to rank them by priority, surface frustrated users, flag competitor mentions, and collect workarounds your community already found. Results arrive sorted high to low, whether you're using the web chat or Slack adapter.\n\n## Usage\n\nSend a message:\n\n| Message | Effect |\n|---------|--------|\n| `owner/repo` | Analyse top 5 open issues |\n| `owner/repo 20` | Analyse top 20 open issues (max 50) |\n| `owner/repo#123` | Analyse a single issue |\n\n**Examples:**\n- *\"pallets/flask\"* — score the 5 most recent open issues in Flask\n- *\"vercel/next.js 15\"* — deep-dive the top 15 Next.js issues\n- *\"rails/rails#50234\"* — analyse one specific Rails issue\n\n## What the report includes\n\nFor each issue:\n- **Priority** — `HIGH`, `MEDIUM`, or `LOW` with a one-sentence justification\n- **Sentiment** — `FRUSTRATION`, `URGENCY`, `NEUTRAL`, or `POSITIVE` with details\n- **Summary** — 2-3 sentence description of the bug or request\n- **Reactions** — upvotes, total reactions, comment count\n- **Competitor mentions** — other tools users compared against (if any)\n- **Workarounds** — solutions users found themselves (if any)\n\nIssues are sorted high to low in the final report.\n\n## Priority definitions\n\n| Priority | Criteria |\n|----------|----------|\n| `HIGH` | Security vulnerability, data loss, crash, or blocker |\n| `MEDIUM` | Significant bug, degraded UX, or important feature request |\n| `LOW` | Minor bug, cosmetic issue, nice-to-have, or question |\n\n## Slack integration\n\nThe agent supports both the **web** and **Slack** adapters. At deploy time, choose which adapter(s) to enable — the Slack adapter handles authentication and message delivery automatically, no extra configuration needed in the agent itself.\n\n## Environment variables\n\n| Variable | Description |\n|----------|-------------|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `GITHUB_TOKEN` | Auto-injected by Astropods GitHub integration |\n\n## Limitations\n\n- Analyses open issues only; closed issues and pull requests are excluded.\n- Analysis is capped at 50 issues per run to stay within token limits.\n- Comment bodies are truncated at 500 characters each; very long comment threads may lose detail.\n- Requires a GitHub token with at least read access to the target repository.\n- The full report is returned once all issues are processed; per-issue progress is not streamed.\n",{"_379":2366,"_381":2367,"_383":2368,"_385":2369,"_387":2370,"_389":2371},"2026-09-28T19:33:48Z","Scores GitHub issues by priority and sentiment using GPT-4o mini, surfacing competitor mentions and workarounds — via web or Slack.",[408,414,516,458,519,402,2380],{"_62":408,"_409":410,"_411":2353},[2374,2375,2376,2377,2378,2379],"\n# GitHub Issue Scorer\n\nYour backlog is full of open issues and you don't know where to start. GitHub Issue Scorer reads every issue — body and all comments — then uses GPT-4o mini to rank them by priority, surface frustrated users, flag competitor mentions, and collect workarounds your community already found. Results arrive sorted high to low, whether you're using the web chat or Slack adapter.\n\n## Usage\n\nSend a message:\n\n| Message | Effect |\n|---------|--------|\n| `owner/repo` | Analyse top 5 open issues |\n| `owner/repo 20` | Analyse top 20 open issues (max 50) |\n| `owner/repo#123` | Analyse a single issue |\n\n**Examples:**\n- *\"pallets/flask\"* — score the 5 most recent open issues in Flask\n- *\"vercel/next.js 15\"* — deep-dive the top 15 Next.js issues\n- *\"rails/rails#50234\"* — analyse one specific Rails issue\n\n## What the report includes\n\nFor each issue:\n- **Priority** — `HIGH`, `MEDIUM`, or `LOW` with a one-sentence justification\n- **Sentiment** — `FRUSTRATION`, `URGENCY`, `NEUTRAL`, or `POSITIVE` with details\n- **Summary** — 2-3 sentence description of the bug or request\n- **Reactions** — upvotes, total reactions, comment count\n- **Competitor mentions** — other tools users compared against (if any)\n- **Workarounds** — solutions users found themselves (if any)\n\nIssues are sorted high to low in the final report.\n\n## Priority definitions\n\n| Priority | Criteria |\n|----------|----------|\n| `HIGH` | Security vulnerability, data loss, crash, or blocker |\n| `MEDIUM` | Significant bug, degraded UX, or important feature request |\n| `LOW` | Minor bug, cosmetic issue, nice-to-have, or question |\n\n## Slack integration\n\nThe agent supports both the **web** and **Slack** adapters. At deploy time, choose which adapter(s) to enable — the Slack adapter handles authentication and message delivery automatically, no extra configuration needed in the agent itself.\n\n## Environment variables\n\n| Variable | Description |\n|----------|-------------|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `GITHUB_TOKEN` | Auto-injected by Astropods GitHub integration |\n\n## Limitations\n\n- Analyses open issues only; closed issues and pull requests are excluded.\n- Analysis is capped at 50 issues per run to stay within token limits.\n- Comment bodies are truncated at 500 characters each; very long comment threads may lose detail.\n- Requires a GitHub token with at least read access to the target repository.\n- The full report is returned once all issues are processed; per-issue progress is not streamed.\n",[2372,2373],{"_59":408,"_61":1337,"_397":24},{"_59":402,"_61":403,"_397":24},"Fetch and analyse up to 50 open GitHub issues in a single run","Score each issue HIGH, MEDIUM, or LOW priority using GPT-4o mini","Detect user sentiment: frustration, urgency, neutral, or positive","Surface competitor mentions and user-reported workarounds","Deliver ranked priority report via web chat or Slack adapter","Analyse a single issue by number (owner/repo#123)","issues",{"_465":2391,"_450":2392},{"_450":2387},{"_408":2386},{"_402":2385},{"_426":402},{"_426":408},{"_454":2388},{"_456":2389,"_458":2390},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/github-issue-scorer:59cc69cd",{"_454":24},"#a379d2","#ca9efa","#a14cff","#572d86","#bd99e6","#b085e0","github-issue-analyzer","https://assets.astropods.ai/avatars/agents/simon/github-issue-analyzer.jpg",{"_468":2457,"_470":2458,"_472":2459,"_474":475,"_476":2460,"_478":2461,"_480":1274,"_482":2462,"_484":2463},[2408],{"_366":2406,"_367":2407},[2405],{"_61":31,"_347":31},17,37,{"_369":2409,"_371":2410,"_373":2411,"_375":2412,"_377":2413},"8dbbf7a9",{"_416":2430,"_418":2431,"_955":2432,"_420":2433,"_389":2434,"_958":2435,"_422":2436,"_371":424},"---\ndescription: Understand what's happening in your GitHub repository — surface issue trends, categories, competitor mentions, and solutions without writing a single query\ntags: [\"github\", \"issues\", \"knowledge-graph\", \"analytics\"]\ncapabilities:\n - \"Query GitHub issues using natural language via a Neo4j knowledge graph\"\n - \"Surface issue trends, categories, competitor mentions, and workarounds\"\n - \"Summarize comments on a specific issue\"\n - \"Run incremental or full syncs of repository issues into the graph\"\nintegrations:\n - \"GitHub\"\n - \"Neo4j\"\nrepository:\n type: git\n url: \"https://github.com/astropods/agents\"\n directory: github-issue-analyzer\n---\n\nReading through hundreds of GitHub issues to spot patterns, find workarounds, or understand what competitors are mentioned takes hours. This agent does it for you — it ingests your repository's issues into a knowledge graph, enriches them with AI analysis, and lets you ask questions in plain English to get instant answers.\n\n## What you can ask\n\n- \"What are the most common categories of open issues?\"\n- \"Are there any issues mentioning competitor X?\"\n- \"Summarize the comments on issue #42\"\n- \"Which issues have no workaround yet?\"\n- \"What bugs have the most reactions?\"\n- \"Show me all issues in the 'performance' category\"\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `queryNeo4j` | Runs read-only Cypher queries against the knowledge graph |\n| `summarizeComments` | Fetches and summarizes comments for a specific issue |\n\n## Knowledge graph\n\nIssues are stored as nodes with rich relationships:\n\n- **Categories** — bug, feature, performance, docs, etc. (AI-assigned)\n- **Competitors** — any competitor tools mentioned in the issue body or comments\n- **Workarounds / Solutions** — extracted by OpenAI and stored as linked nodes\n- **Keywords** — salient terms for search and clustering\n- **Users** — authors and commenters\n\n## Ingestion\n\nRuns automatically on startup. Set `SYNC_MODE=schedule` for incremental syncs (only fetches issues updated since the last run). The target repo is set via the top-level `GITHUB_OWNER` / `GITHUB_REPO` inputs in `astropods.yml` (configured at deploy time with `ast configure`); tune `ISSUE_LIMIT` and `SYNC_MODE` build args in `astropods.yml`.\n",{"_379":2414,"_381":2415,"_383":2416,"_385":2417,"_387":2418,"_389":2419},"2026-09-28T19:33:55Z","Understand what's happening in your GitHub repository — surface issue trends, categories, competitor mentions, and solutions without writing a single query",[408,2380,2428,2429],{"_62":574,"_409":410,"_411":2399},[2424,2425,2426,2427],"\nReading through hundreds of GitHub issues to spot patterns, find workarounds, or understand what competitors are mentioned takes hours. This agent does it for you — it ingests your repository's issues into a knowledge graph, enriches them with AI analysis, and lets you ask questions in plain English to get instant answers.\n\n## What you can ask\n\n- \"What are the most common categories of open issues?\"\n- \"Are there any issues mentioning competitor X?\"\n- \"Summarize the comments on issue #42\"\n- \"Which issues have no workaround yet?\"\n- \"What bugs have the most reactions?\"\n- \"Show me all issues in the 'performance' category\"\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `queryNeo4j` | Runs read-only Cypher queries against the knowledge graph |\n| `summarizeComments` | Fetches and summarizes comments for a specific issue |\n\n## Knowledge graph\n\nIssues are stored as nodes with rich relationships:\n\n- **Categories** — bug, feature, performance, docs, etc. (AI-assigned)\n- **Competitors** — any competitor tools mentioned in the issue body or comments\n- **Workarounds / Solutions** — extracted by OpenAI and stored as linked nodes\n- **Keywords** — salient terms for search and clustering\n- **Users** — authors and commenters\n\n## Ingestion\n\nRuns automatically on startup. Set `SYNC_MODE=schedule` for incremental syncs (only fetches issues updated since the last run). The target repo is set via the top-level `GITHUB_OWNER` / `GITHUB_REPO` inputs in `astropods.yml` (configured at deploy time with `ast configure`); tune `ISSUE_LIMIT` and `SYNC_MODE` build args in `astropods.yml`.\n",[2420,2421],{"_59":408,"_61":1337,"_397":24},{"_59":2422,"_61":2423,"_397":242},"neo4j","Neo4j","Query GitHub issues using natural language via a Neo4j knowledge graph","Surface issue trends, categories, competitor mentions, and workarounds","Summarize comments on a specific issue","Run incremental or full syncs of repository issues into the graph","knowledge-graph","analytics",{"_465":2456},{"_450":2455},{"_1015":2447,"_1201":2448},{"_2441":2442,"_2443":2444},{"_408":2440},{"_2438":2439},{"_402":2437},{"_426":402},"graph",{"_426":2422},{"_426":408},"GITHUB_OWNER",{"_439":440,"_379":2446,"_61":2441},"GITHUB_REPO",{"_439":440,"_379":2445,"_61":2443},"Name of the GitHub repo to ingest issues from","Owner of the GitHub repo to ingest issues from",{"_1010":2452,"_1013":2453},{"_1010":2449,"_1013":2450},{"_465":2451},{"_62":1201},"registry.astropods.ai/01b3f44b-247a-4a6e-8328-ebe90829086e/github-issue-analyzer-ingestion-startup:8dbbf7a9",{"_465":2454},{"_62":1015},"registry.astropods.ai/01b3f44b-247a-4a6e-8328-ebe90829086e/github-issue-analyzer-ingestion-schedule:8dbbf7a9",[462],"registry.astropods.ai/01b3f44b-247a-4a6e-8328-ebe90829086e/github-issue-analyzer:8dbbf7a9","#a48637","#fae19e","#db9f00","#866d2d","#221c0b","#d9cba6","#e0c785","feature-flag-assistant","https://assets.astropods.ai/avatars/agents/astropods/feature-flag-assistant.jpg",{"_468":2533,"_470":2534,"_472":2535,"_474":475,"_476":1211,"_478":1212,"_480":875,"_482":2536,"_484":1838},[2473],{"_366":1456,"_367":1130},[2470],{"_61":2471,"_347":2472},"Jessye Coleman-Shapiro","jessyecs",{"_369":2474,"_371":2475,"_373":2476,"_375":2477,"_377":2478},"9e09e981",{"_416":2496,"_418":2497,"_955":2498,"_420":2499,"_389":2500,"_422":2501,"_371":424},"---\ndescription: \"Identifies feature flag code that can be removed and old feature flags to be deleted\"\ntags: [\"feature-flags\", \"launchdarkly\", \"cleanup\"]\ncapabilities:\n - \"Identify LaunchDarkly flags fully rolled out to 100% in production for 2+ weeks\"\n - \"Identify LaunchDarkly flags with zero code references\"\n - \"Search GitHub for code references to a specific flag\"\n - \"Preview or manually send bi-weekly scheduled flag audit reports to Slack\"\nintegrations:\n - \"LaunchDarkly\"\n - \"GitHub\"\n - \"Slack\"\nrepository:\n type: git\n url: \"https://github.com/astropods/agents\"\n directory: feature-flag-assistant\n---\n\n# Feature Flag Helper Agent\n\nHelps engineering teams identify stale LaunchDarkly feature flags for cleanup.\n\n## Capabilities\n\n### Fully Rolled Out Flags\nFlags that have been serving 100% of production traffic to their \"on\" variation for 2+ weeks.\nThe surrounding conditional code can be deleted and the \"on\" path hardcoded.\n\n### Orphaned Flags (No Code References)\nFlags that appear in LaunchDarkly but have zero references in any scanned repository.\nThese can be deleted directly from LaunchDarkly.\n\n### Scheduled Slack Reports\nA bi-weekly audit report is automatically posted to a configured Slack channel. The report highlights flags that newly crossed the 2-week rollout threshold and the 5 oldest flags still awaiting cleanup. The report schedule is configured at deploy time via the UI.\n\n## Required Configuration\n\nSet the following secrets before running:\n\n| Variable | Required | Description |\n|---|---|---|\n| `LAUNCHDARKLY_API_KEY` | Yes | LD REST API key with `reader` role |\n| `LAUNCHDARKLY_PROJECT_KEY` | No | LD project key (default: `default`) |\n| `LAUNCHDARKLY_PRODUCTION_ENV` | No | LD production environment key (default: `production`) |\n| `GITHUB_REPO` | No | Repository to search for flag code references (`owner/repo`) |\n| `SLACK_NOTIFY_CHANNEL` | No | Slack channel ID or name to post scheduled audit reports to |\n\nThe LaunchDarkly API key needs:\n- `reader` access to Flags\n- `reader` access to Code References (for the no-code-refs tool)\n\nCode References must be enabled in your LaunchDarkly project and at least one repository\nscan must have run for the orphaned-flag tool to return data.\n\n## Example Prompts\n\n- \"Which feature flags can I clean up?\"\n- \"Show me flags that have been fully rolled out to production for over 2 weeks\"\n- \"List all flags with no code references\"\n- \"Give me details on the flag `my-flag-key`\"\n",{"_379":2479,"_381":2480,"_383":2481,"_385":2482,"_387":2483,"_389":2484},"2026-09-28T19:33:49Z","Identifies feature flag code that can be removed and old feature flags to be deleted",[2494,2488,2495],{"_62":574,"_409":410,"_411":2464},[2490,2491,2492,2493],"\n# Feature Flag Helper Agent\n\nHelps engineering teams identify stale LaunchDarkly feature flags for cleanup.\n\n## Capabilities\n\n### Fully Rolled Out Flags\nFlags that have been serving 100% of production traffic to their \"on\" variation for 2+ weeks.\nThe surrounding conditional code can be deleted and the \"on\" path hardcoded.\n\n### Orphaned Flags (No Code References)\nFlags that appear in LaunchDarkly but have zero references in any scanned repository.\nThese can be deleted directly from LaunchDarkly.\n\n### Scheduled Slack Reports\nA bi-weekly audit report is automatically posted to a configured Slack channel. The report highlights flags that newly crossed the 2-week rollout threshold and the 5 oldest flags still awaiting cleanup. The report schedule is configured at deploy time via the UI.\n\n## Required Configuration\n\nSet the following secrets before running:\n\n| Variable | Required | Description |\n|---|---|---|\n| `LAUNCHDARKLY_API_KEY` | Yes | LD REST API key with `reader` role |\n| `LAUNCHDARKLY_PROJECT_KEY` | No | LD project key (default: `default`) |\n| `LAUNCHDARKLY_PRODUCTION_ENV` | No | LD production environment key (default: `production`) |\n| `GITHUB_REPO` | No | Repository to search for flag code references (`owner/repo`) |\n| `SLACK_NOTIFY_CHANNEL` | No | Slack channel ID or name to post scheduled audit reports to |\n\nThe LaunchDarkly API key needs:\n- `reader` access to Flags\n- `reader` access to Code References (for the no-code-refs tool)\n\nCode References must be enabled in your LaunchDarkly project and at least one repository\nscan must have run for the orphaned-flag tool to return data.\n\n## Example Prompts\n\n- \"Which feature flags can I clean up?\"\n- \"Show me flags that have been fully rolled out to production for over 2 weeks\"\n- \"List all flags with no code references\"\n- \"Give me details on the flag `my-flag-key`\"\n",[2485,2486,2487],{"_59":2488,"_61":2489,"_397":242},{"_59":408,"_61":1337,"_397":24},{"_59":458,"_61":1083,"_397":24},"launchdarkly","LaunchDarkly","Identify LaunchDarkly flags fully rolled out to 100% in production for 2+ weeks","Identify LaunchDarkly flags with zero code references","Search GitHub for code references to a specific flag","Preview or manually send bi-weekly scheduled flag audit reports to Slack","feature-flags","cleanup",{"_465":2532},{"_450":2529},{"_2524":2525},{"_2504":2505,"_2506":2507,"_2508":2509,"_2510":2511,"_2512":2513},{"_408":2503},{"_568":2502},{"_426":568},{"_426":408},"github_repo",{"_439":440,"_379":2523,"_61":2443},"launchdarkly_api_key",{"_439":440,"_379":2521,"_61":2522,"_445":24},"launchdarkly_production_env",{"_439":440,"_642":2518,"_379":2519,"_61":2520},"launchdarkly_project_key",{"_439":440,"_642":642,"_379":2516,"_61":2517},"slack_notify_channel",{"_439":440,"_379":2514,"_61":2515},"Slack channel ID or name to post scheduled flag audit reports to","SLACK_NOTIFY_CHANNEL","LaunchDarkly project key (e.g. \"default\" or your project slug)","LAUNCHDARKLY_PROJECT_KEY","production","LaunchDarkly environment key that represents production","LAUNCHDARKLY_PRODUCTION_ENV","LaunchDarkly REST API key with read access to flags and code references","LAUNCHDARKLY_API_KEY","GitHub repository to search for flag references (format: owner/repo)","scheduled-report",{"_1010":2526,"_1013":2527},{"_465":2528},{"_62":1015},"registry.astropods.ai/cd210ffd-25bf-4acb-9158-da91e59a883a/feature-flag-assistant-ingestion-scheduled-report:9e09e981",{"_454":2530},{"_456":2531},[462,458],"registry.astropods.ai/cd210ffd-25bf-4acb-9158-da91e59a883a/feature-flag-assistant:9e09e981","#653739","#eaaeb0","#7c2023","#d9a6a7","droplit","https://assets.astropods.ai/avatars/agents/rabbah/droplit.jpg",{"_468":2588,"_470":2589,"_472":2590,"_474":475,"_476":2591,"_478":2592,"_480":2279,"_482":2593,"_484":2594},[2544],{"_366":1130,"_367":1721},[2543],{"_61":192,"_347":37},{"_369":2545,"_371":2546,"_373":2547,"_375":2548,"_377":2549},"1c783a51",{"_416":2572,"_418":2573,"_958":2574,"_862":2575,"_371":584},"---\ndescription: \"A tiny single-domain host: drop an HTML or markdown file or a whole folder into the admin UI and it is published at // with a Claude-derived slug, title, and description.\"\ntags:\n - hosting\n - html\n - markdown\nauthors:\n - name: Rodric Rabbah\n account: rabbah\ncapabilities:\n - Gates the /admin UI to an email allowlist matched against the platform OIDC identity\n - Accepts dropped HTML/markdown files or whole folders and publishes each at //\n - Serves a dropped folder as a multi-file site, with binary assets and correct content types\n - Calls Claude on upload to derive a URL slug, title, and one-line description from the contents\n - Stores single pages and site bundles in Redis and refreshes its in-process cache via pub/sub\n - Serves a styled 404 page for unknown paths and missing site assets\n - Exposes a list_pages tool the chat agent uses to enumerate what is hosted\nrepository: github:rabbah/droplit\nintegrations:\n - Anthropic\n - Redis\n - Mastra\n---\n\n## Overview\n\n`droplit` is a tiny single-domain hosting agent. Drop an HTML or markdown file,\nor a whole folder, into the admin UI. On each upload Claude derives a short URL\nslug (for example `trip-itinerary`), a page title, and a one-line description\nfrom the contents. A single file becomes a page; a folder becomes a multi-file\nsite served at `//`, with its `index.html` as the entry point and every\nasset reachable beneath it. Markdown is converted to HTML at upload time and\nwrapped in a minimal styled template.\n\nThe chat agent's only job is to help visitors discover what is hosted. It calls\nthe `list_pages` tool to fetch the current catalog and replies with the URL,\ntitle, and one-line description for each entry.\n\n## Usage\n\n### Publishing pages and sites\n\nAccess to `/admin` is granted by email. The platform signs you in through its\nOpenID Connect (OIDC) layer, and your verified email must appear in the\n`ADMIN_EMAILS` allowlist. In local development there is no sign-in, so the admin\nUI is open. Drag and drop files or a folder into the upload area:\n\n- **A single file** (`.html`, `.htm`, `.md`, or `.markdown`) becomes a page at\n `//`.\n- **A folder** becomes a site. Its files are served at `//` with the\n right content type, binary assets included. If the folder has no `index.html`,\n a file listing is generated as the entry point.\n\nFor each upload Claude picks a kebab-case slug, a title, and a one-line\ndescription. If a slug collides with an existing entry, a numeric suffix is\nappended (`my-page` becomes `my-page-2`). Pages and sites are stored in Redis\nand reachable immediately; changes propagate to the agent's in-process cache via\npub/sub with no restart. Existing entries can be deleted from the same view.\n\nA bundled `welcome` page is seeded on first boot if Redis is empty.\n\n### Browsing\n\n- `GET //` serves a page, or a site's entry document.\n- `GET //` serves an asset inside a site; a trailing slash resolves\n to that directory's `index.html`.\n- `GET /` redirects to `//` so relative links resolve.\n- Unknown paths and missing assets return a styled 404 page.\n- Reserved prefixes (`admin`, `api`, `healthz`, `static`, `assets`, `whoami`)\n cannot be used as slugs.\n\nEvery route other than `/admin` is open to any visitor.\n\n### Asking the chat agent what is available\n\nIn any messaging adapter (web playground, Slack, and so on), ask the agent what\nis hosted. For example, ask it *\"what pages are available here?\"* It calls\n`list_pages` and replies with bulleted entries that include the path, title, and\ndescription. It reports only pages that exist and does not invent them.\n\n### Inputs\n\n| Input | Required | Description |\n|---|---|---|\n| `ADMIN_EMAILS` | yes | Comma-separated admin emails allowed at `/admin`, matched case-insensitively against the OIDC email |\n\n## Limitations\n\n- **Admin access needs the platform identity.** In production the `/admin` gate\n reads the verified email from the platform's OIDC token, so `ADMIN_EMAILS`\n must be set or no one can manage pages. There is no password fallback.\n- **Per-file uploads are capped at 1 MiB.** A dropped folder is additionally\n capped at 200 files and 20 MiB total, and is rejected as a whole if it exceeds\n either limit.\n- **Page and site content is served as-is.** The admin uploads HTML and assets\n directly and the host does not sanitize them, so add only people you trust to\n `ADMIN_EMAILS`.\n- **Multi-replica safe.** Pages and site bundles live in Redis, so any replica\n can serve any entry, and the in-process metadata cache is refreshed via\n pub/sub.\n",{"_379":2550,"_381":2551,"_559":2552,"_383":2553,"_385":2554,"_387":2555,"_389":2556},"2026-06-24T12:17:40Z","A tiny single-domain host: drop an HTML or markdown file or a whole folder into the admin UI and it is published at // with a Claude-derived slug, title, and description.",[2569,2570,2571],[2568],{"_62":574,"_409":2567},[2560,2561,2562,2563,2564,2565,2566],"\n## Overview\n\n`droplit` is a tiny single-domain hosting agent. Drop an HTML or markdown file,\nor a whole folder, into the admin UI. On each upload Claude derives a short URL\nslug (for example `trip-itinerary`), a page title, and a one-line description\nfrom the contents. A single file becomes a page; a folder becomes a multi-file\nsite served at `//`, with its `index.html` as the entry point and every\nasset reachable beneath it. Markdown is converted to HTML at upload time and\nwrapped in a minimal styled template.\n\nThe chat agent's only job is to help visitors discover what is hosted. It calls\nthe `list_pages` tool to fetch the current catalog and replies with the URL,\ntitle, and one-line description for each entry.\n\n## Usage\n\n### Publishing pages and sites\n\nAccess to `/admin` is granted by email. The platform signs you in through its\nOpenID Connect (OIDC) layer, and your verified email must appear in the\n`ADMIN_EMAILS` allowlist. In local development there is no sign-in, so the admin\nUI is open. Drag and drop files or a folder into the upload area:\n\n- **A single file** (`.html`, `.htm`, `.md`, or `.markdown`) becomes a page at\n `//`.\n- **A folder** becomes a site. Its files are served at `//` with the\n right content type, binary assets included. If the folder has no `index.html`,\n a file listing is generated as the entry point.\n\nFor each upload Claude picks a kebab-case slug, a title, and a one-line\ndescription. If a slug collides with an existing entry, a numeric suffix is\nappended (`my-page` becomes `my-page-2`). Pages and sites are stored in Redis\nand reachable immediately; changes propagate to the agent's in-process cache via\npub/sub with no restart. Existing entries can be deleted from the same view.\n\nA bundled `welcome` page is seeded on first boot if Redis is empty.\n\n### Browsing\n\n- `GET //` serves a page, or a site's entry document.\n- `GET //` serves an asset inside a site; a trailing slash resolves\n to that directory's `index.html`.\n- `GET /` redirects to `//` so relative links resolve.\n- Unknown paths and missing assets return a styled 404 page.\n- Reserved prefixes (`admin`, `api`, `healthz`, `static`, `assets`, `whoami`)\n cannot be used as slugs.\n\nEvery route other than `/admin` is open to any visitor.\n\n### Asking the chat agent what is available\n\nIn any messaging adapter (web playground, Slack, and so on), ask the agent what\nis hosted. For example, ask it *\"what pages are available here?\"* It calls\n`list_pages` and replies with bulleted entries that include the path, title, and\ndescription. It reports only pages that exist and does not invent them.\n\n### Inputs\n\n| Input | Required | Description |\n|---|---|---|\n| `ADMIN_EMAILS` | yes | Comma-separated admin emails allowed at `/admin`, matched case-insensitively against the OIDC email |\n\n## Limitations\n\n- **Admin access needs the platform identity.** In production the `/admin` gate\n reads the verified email from the platform's OIDC token, so `ADMIN_EMAILS`\n must be set or no one can manage pages. There is no password fallback.\n- **Per-file uploads are capped at 1 MiB.** A dropped folder is additionally\n capped at 200 files and 20 MiB total, and is rejected as a whole if it exceeds\n either limit.\n- **Page and site content is served as-is.** The admin uploads HTML and assets\n directly and the host does not sanitize them, so add only people you trust to\n `ADMIN_EMAILS`.\n- **Multi-replica safe.** Pages and site bundles live in Redis, so any replica\n can serve any entry, and the in-process metadata cache is refreshed via\n pub/sub.\n",[2557,2558,2559],{"_59":568,"_61":569,"_397":24},{"_59":1084,"_61":1085,"_397":24},{"_59":579,"_61":1082,"_397":242},"Gates the /admin UI to an email allowlist matched against the platform OIDC identity","Accepts dropped HTML/markdown files or whole folders and publishes each at //","Serves a dropped folder as a multi-file site, with binary assets and correct content types","Calls Claude on upload to derive a URL slug, title, and one-line description from the contents","Stores single pages and site bundles in Redis and refreshes its in-process cache via pub/sub","Serves a styled 404 page for unknown paths and missing site assets","Exposes a list_pages tool the chat agent uses to enumerate what is hosted","https://github.com/rabbah/droplit",{"_61":192,"_347":37},"hosting","html","markdown",{"_1427":24,"_465":2582,"_420":2583,"_450":2584},{"_450":2577},{"_1084":2576},{},{"_426":1084},{"_452":2578,"_454":2579},{"_463":1113},{"_456":2580,"_458":2581},[462],{"_460":24,"_461":24},"registry.astropods.ai/e9945c0f-c341-421f-8680-473e88c1807c/droplit:1c783a51",[2585],{"_452":24,"_454":24},{"_439":440,"_379":2586,"_61":2587},"Comma-separated list of admin email addresses allowed to access /admin (matched against the OIDC email, case-insensitively)","ADMIN_EMAILS","#73a9a8","#b4e4e3","#58c4c1","#2f8482","#0c2221","#a6d9d7","#8ed7d5","daily-customer-briefs","https://assets.astropods.ai/avatars/agents/astro-ai/daily-customer-briefs.jpg",{"_468":2667,"_470":2668,"_472":2669,"_474":475,"_476":2670,"_478":1314,"_480":875,"_482":2671,"_484":1316},[2602],{"_366":303,"_367":303},[2601],{"_61":192,"_347":41},{"_369":2603,"_371":2604,"_373":2605,"_375":2606,"_377":2607},"fe02d0dd",{"_416":2628,"_418":2629,"_420":2630,"_422":2631,"_371":424},"---\ndescription: \"Daily pre-meeting brief agent that fetches Google Calendar events, looks up Zendesk tickets and HubSpot deals per attendee, and posts a customer brief to Slack each weekday morning.\"\ntags:\n - google-calendar\n - zendesk\n - hubspot\n - slack\n - openai\n - cron\ncapabilities:\n - \"Fetch today's Google Calendar events\"\n - \"Look up open Zendesk tickets per meeting attendee\"\n - \"Look up active HubSpot deals per meeting attendee\"\n - \"Post a formatted pre-meeting customer brief to Slack on a cron schedule\"\nrepository:\n type: github\n url: https://github.com/astropods/agents\n directory: customer-meetings-agent\nintegrations:\n - OpenAI\n - Google Calendar\n - Zendesk\n - HubSpot\n - Slack (via built-in adapter)\n---\n\n# Customer Meetings Agent\n\nRuns every weekday morning and posts a pre-meeting brief to Slack. For each calendar event with external attendees, the agent looks up open Zendesk tickets and active HubSpot deals for those customers and formats a concise per-meeting summary.\n\n## Triggers\n\n- **Cron job** (default `0 8 * * 1-5`): Automatically generates and posts the brief each weekday at 8 AM. Override with the `CRON_SCHEDULE` env var.\n- **Chat message**: Request a brief on demand via the web or Slack adapter.\n\n## Tools\n\n| Tool | Description |\n|---|---|\n| `get_calendar_events` | Fetches events from Google Calendar for a given date (defaults to today) |\n| `get_zendesk_tickets` | Searches Zendesk for open tickets matching a customer name or email |\n| `get_hubspot_deals` | Searches HubSpot for active deals matching a company name |\n\n## Environment variables\n\n| Variable | Description |\n|---|---|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `GOOGLE_CLIENT_ID` | Google OAuth client ID |\n| `GOOGLE_CLIENT_SECRET` | Google OAuth client secret |\n| `GOOGLE_REFRESH_TOKEN` | Google OAuth refresh token for Calendar access |\n| `GOOGLE_CALENDAR_ID` | Calendar ID to read events from (`primary` or a calendar email address) |\n| `ZENDESK_URL` | Zendesk instance URL (e.g. `https://yourcompany.zendesk.com`) |\n| `ZENDESK_EMAIL` | Zendesk agent email address (treated as a secret) |\n| `ZENDESK_API_KEY` | Zendesk API token |\n| `HUBSPOT_API_KEY` | HubSpot private app access token |\n| `SLACK_CHANNEL` | Slack channel ID for the daily brief (optional) |\n| `CRON_SCHEDULE` | Cron expression (default: `0 8 * * 1-5`) |\n",{"_379":2608,"_381":2609,"_383":2610,"_385":2611,"_387":2612,"_389":2613},"2026-09-28T19:35:32Z","Daily pre-meeting brief agent that fetches Google Calendar events, looks up Zendesk tickets and HubSpot deals per attendee, and posts a customer brief to Slack each weekday morning.",[2621,400,2619,458,402,1099],{"_62":408,"_409":410,"_411":2627},[2623,2624,2625,2626],"\n# Customer Meetings Agent\n\nRuns every weekday morning and posts a pre-meeting brief to Slack. For each calendar event with external attendees, the agent looks up open Zendesk tickets and active HubSpot deals for those customers and formats a concise per-meeting summary.\n\n## Triggers\n\n- **Cron job** (default `0 8 * * 1-5`): Automatically generates and posts the brief each weekday at 8 AM. Override with the `CRON_SCHEDULE` env var.\n- **Chat message**: Request a brief on demand via the web or Slack adapter.\n\n## Tools\n\n| Tool | Description |\n|---|---|\n| `get_calendar_events` | Fetches events from Google Calendar for a given date (defaults to today) |\n| `get_zendesk_tickets` | Searches Zendesk for open tickets matching a customer name or email |\n| `get_hubspot_deals` | Searches HubSpot for active deals matching a company name |\n\n## Environment variables\n\n| Variable | Description |\n|---|---|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `GOOGLE_CLIENT_ID` | Google OAuth client ID |\n| `GOOGLE_CLIENT_SECRET` | Google OAuth client secret |\n| `GOOGLE_REFRESH_TOKEN` | Google OAuth refresh token for Calendar access |\n| `GOOGLE_CALENDAR_ID` | Calendar ID to read events from (`primary` or a calendar email address) |\n| `ZENDESK_URL` | Zendesk instance URL (e.g. `https://yourcompany.zendesk.com`) |\n| `ZENDESK_EMAIL` | Zendesk agent email address (treated as a secret) |\n| `ZENDESK_API_KEY` | Zendesk API token |\n| `HUBSPOT_API_KEY` | HubSpot private app access token |\n| `SLACK_CHANNEL` | Slack channel ID for the daily brief (optional) |\n| `CRON_SCHEDULE` | Cron expression (default: `0 8 * * 1-5`) |\n",[2614,2615,2616,2617,2618],{"_59":402,"_61":403,"_397":24},{"_59":2621,"_61":2622,"_397":24},{"_59":400,"_61":401,"_397":24},{"_59":2619,"_61":2620,"_397":24},{"_59":395,"_61":396,"_397":242},"hubspot","HubSpot","google-calendar","Google Calendar","Fetch today's Google Calendar events","Look up open Zendesk tickets per meeting attendee","Look up active HubSpot deals per meeting attendee","Post a formatted pre-meeting customer brief to Slack on a cron schedule","customer-meetings-agent",{"_465":2665,"_450":2666},{"_450":2661},{"_2633":2634,"_2635":2636,"_2637":2638,"_2639":2640,"_2641":2642,"_2643":2644,"_1830":2645,"_435":2646,"_2647":2648,"_2649":2650},{"_402":2632},{"_426":402},"CRON_SCHEDULE",{"_439":440,"_642":2659,"_379":2660,"_442":443,"_61":2633,"_984":24},"GOOGLE_CALENDAR_ID",{"_439":440,"_379":2658,"_442":443,"_61":2635},"GOOGLE_CLIENT_ID",{"_439":440,"_379":2657,"_442":443,"_61":2637,"_445":24},"GOOGLE_CLIENT_SECRET",{"_439":440,"_379":2656,"_442":443,"_61":2639,"_445":24},"GOOGLE_REFRESH_TOKEN",{"_439":440,"_379":2655,"_442":443,"_61":2641,"_445":24},"HUBSPOT_API_KEY",{"_439":440,"_379":2654,"_442":443,"_61":2643,"_445":24},{"_439":440,"_379":2653,"_442":443,"_61":1830,"_984":24},{"_439":440,"_379":444,"_442":443,"_61":435,"_445":24},"ZENDESK_EMAIL",{"_439":440,"_379":2652,"_442":443,"_61":2647,"_445":24},"ZENDESK_URL",{"_439":440,"_379":2651,"_442":443,"_61":2649},"Zendesk instance URL (e.g. https://yourcompany.zendesk.com)","Zendesk agent email address","Slack channel ID to post the daily brief to","HubSpot private app access token","Google OAuth refresh token for Calendar access","Google OAuth client secret","Google OAuth client ID","Google Calendar ID to read events from (e.g. primary or a calendar email)","0 8 * * 1-5","Cron expression for the daily brief schedule",{"_454":2662},{"_456":2663,"_458":2664},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/daily-customer-briefs:fe02d0dd",{"_454":24},"#8e5a4d","#eabaae","#ae472d","#863f2d","#e6999b","customer-call-followup","https://assets.astropods.ai/avatars/agents/astro-ai/customer-call-followup.jpg",{"_468":2741,"_470":871,"_472":2742,"_474":475,"_476":873,"_478":874,"_480":875,"_482":876,"_484":877},[2679],{"_366":303,"_367":303},[2678],{"_61":192,"_347":41},{"_369":2680,"_371":2681,"_373":2682,"_375":2683,"_377":2684},"fcbe4509",{"_416":2709,"_418":2710,"_420":2711,"_422":2712,"_371":424},"---\ndescription: \"Fetches a Zoom call transcript, extracts action items, creates Zendesk tickets for issues, and writes a call summary to Notion.\"\ntags:\n - zoom\n - zendesk\n - notion\n - sales\n - customer-success\n - openai\n - follow-up\ncapabilities:\n - \"Accept a meeting ID via webhook POST, chat, or Slack — extracts it from natural language automatically\"\n - \"Fetch a Zoom meeting transcript via OAuth2\"\n - \"Extract action items from the call, prepended with the account name\"\n - \"Create Zendesk support tickets for issues identified in the call\"\n - \"Create a Notion page with the call summary and action items\"\n - \"Return a clear summary with action items, ticket links, and Notion page link\"\n - \"Report a friendly error when the transcript is still processing\"\nrepository:\n type: github\n url: https://github.com/astropods/agents\n directory: customer-call-follow-up-agent\nintegrations:\n - OpenAI\n - Zoom\n - Zendesk\n - Notion\n - Slack (via built-in adapter)\n---\n\n# Customer Call Follow-Up Agent\n\nYour call just ended and the action items are still in your head. Customer Call Follow-Up Agent reads the Zoom transcript, pulls out every commitment and next step, raises Zendesk tickets for any issues that came up, and drops a clean summary into Notion — then posts the formatted action list wherever you're working.\n\n## Usage\n\n### Chat or Slack\n\nSend a message containing the Zoom meeting ID — it can be a plain ID or natural language:\n\n> *\"87654321098\"*\n> *\"Can you process my call with Acme from this morning — meeting 876 5432 1098\"*\n\nThe agent extracts and validates the meeting ID (9–11 digit number) from your message automatically.\n\n### Webhook\n\nTrigger the agent programmatically by posting the meeting ID explicitly (e.g. from a Zoom recording-ready automation):\n\n```bash\ncurl -X POST https:// \\\n -H \"Content-Type: application/json\" \\\n -d '{\"meetingId\": \"87654321098\"}'\n```\n\nIf `WEBHOOK_SECRET` is set, include the bearer token:\n\n```bash\ncurl -X POST https:// \\\n -H \"Content-Type: application/json\" \\\n -H \"Authorization: Bearer \" \\\n -d '{\"meetingId\": \"87654321098\"}'\n```\n\nIf the transcript is still processing, the agent will respond with a clear message to try again in a few minutes.\n\n## What the response includes\n\n- **Action items** — numbered list, each prepended with the account name\n- **Zendesk tickets** — ID and link for any support issue raised during the call\n- **Notion page** — link to the newly created call summary page\n\n## Tools\n\n| Tool | Description |\n|---|---|\n| `extract_meeting_id` | Extracts and validates a Zoom meeting ID (9–11 digit number) from free-form text |\n| `get_zoom_transcript` | Refreshes OAuth token and downloads the VTT transcript for the meeting |\n| `create_zendesk_ticket` | Creates a support ticket for issues found in the call |\n| `update_notion_page` | Creates a Notion child page with the call summary and action items |\n\n## Environment variables\n\n| Variable | Description |\n|---|---|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `ZOOM_CLIENT_ID` | Zoom marketplace app client ID |\n| `ZOOM_CLIENT_SECRET` | Zoom marketplace app client secret |\n| `ZOOM_REFRESH_TOKEN` | OAuth2 refresh token — obtained by completing Zoom OAuth authorization |\n| `ZENDESK_SUBDOMAIN` | Zendesk subdomain (the `{subdomain}` in `https://{subdomain}.zendesk.com`) |\n| `ZENDESK_AGENT_EMAIL` | Zendesk agent email for API auth |\n| `ZENDESK_API_KEY` | Zendesk API token |\n| `NOTION_API_KEY` | Notion internal integration secret — from [notion.so/my-integrations](https://www.notion.so/my-integrations) |\n| `NOTION_PARENT_PAGE_ID` | ID of the Notion page that holds all call summary pages |\n| `WEBHOOK_SECRET` | Bearer token to authenticate incoming webhook requests — required; all requests without a valid `Authorization: Bearer ` header are rejected with 401 |\n\n## Limitations\n\n- Meeting IDs must be 9–11 digit numbers (standard Zoom format). Vanity meeting URLs are not supported.\n- Fetches cloud recordings only; local Zoom recordings are not supported.\n- Requires a completed transcript (`status: completed`) — processing recordings return a friendly error; retry once Zoom finishes processing.\n- Zendesk tickets are created with the agent email as requester; the sales rep is not automatically assigned unless configured in Zendesk automation rules.\n- Notion output is plain text paragraphs; rich formatting (tables, checkboxes) is not used.\n",{"_379":2685,"_381":2686,"_383":2687,"_385":2688,"_387":2689,"_389":2690},"2026-09-28T19:35:54Z","Fetches a Zoom call transcript, extracts action items, creates Zendesk tickets for issues, and writes a call summary to Notion.",[2696,400,800,2706,2707,402,2708],{"_62":408,"_409":410,"_411":2705},[2698,2699,2700,2701,2702,2703,2704],"\n# Customer Call Follow-Up Agent\n\nYour call just ended and the action items are still in your head. Customer Call Follow-Up Agent reads the Zoom transcript, pulls out every commitment and next step, raises Zendesk tickets for any issues that came up, and drops a clean summary into Notion — then posts the formatted action list wherever you're working.\n\n## Usage\n\n### Chat or Slack\n\nSend a message containing the Zoom meeting ID — it can be a plain ID or natural language:\n\n> *\"87654321098\"*\n> *\"Can you process my call with Acme from this morning — meeting 876 5432 1098\"*\n\nThe agent extracts and validates the meeting ID (9–11 digit number) from your message automatically.\n\n### Webhook\n\nTrigger the agent programmatically by posting the meeting ID explicitly (e.g. from a Zoom recording-ready automation):\n\n```bash\ncurl -X POST https:// \\\n -H \"Content-Type: application/json\" \\\n -d '{\"meetingId\": \"87654321098\"}'\n```\n\nIf `WEBHOOK_SECRET` is set, include the bearer token:\n\n```bash\ncurl -X POST https:// \\\n -H \"Content-Type: application/json\" \\\n -H \"Authorization: Bearer \" \\\n -d '{\"meetingId\": \"87654321098\"}'\n```\n\nIf the transcript is still processing, the agent will respond with a clear message to try again in a few minutes.\n\n## What the response includes\n\n- **Action items** — numbered list, each prepended with the account name\n- **Zendesk tickets** — ID and link for any support issue raised during the call\n- **Notion page** — link to the newly created call summary page\n\n## Tools\n\n| Tool | Description |\n|---|---|\n| `extract_meeting_id` | Extracts and validates a Zoom meeting ID (9–11 digit number) from free-form text |\n| `get_zoom_transcript` | Refreshes OAuth token and downloads the VTT transcript for the meeting |\n| `create_zendesk_ticket` | Creates a support ticket for issues found in the call |\n| `update_notion_page` | Creates a Notion child page with the call summary and action items |\n\n## Environment variables\n\n| Variable | Description |\n|---|---|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `ZOOM_CLIENT_ID` | Zoom marketplace app client ID |\n| `ZOOM_CLIENT_SECRET` | Zoom marketplace app client secret |\n| `ZOOM_REFRESH_TOKEN` | OAuth2 refresh token — obtained by completing Zoom OAuth authorization |\n| `ZENDESK_SUBDOMAIN` | Zendesk subdomain (the `{subdomain}` in `https://{subdomain}.zendesk.com`) |\n| `ZENDESK_AGENT_EMAIL` | Zendesk agent email for API auth |\n| `ZENDESK_API_KEY` | Zendesk API token |\n| `NOTION_API_KEY` | Notion internal integration secret — from [notion.so/my-integrations](https://www.notion.so/my-integrations) |\n| `NOTION_PARENT_PAGE_ID` | ID of the Notion page that holds all call summary pages |\n| `WEBHOOK_SECRET` | Bearer token to authenticate incoming webhook requests — required; all requests without a valid `Authorization: Bearer ` header are rejected with 401 |\n\n## Limitations\n\n- Meeting IDs must be 9–11 digit numbers (standard Zoom format). Vanity meeting URLs are not supported.\n- Fetches cloud recordings only; local Zoom recordings are not supported.\n- Requires a completed transcript (`status: completed`) — processing recordings return a friendly error; retry once Zoom finishes processing.\n- Zendesk tickets are created with the agent email as requester; the sales rep is not automatically assigned unless configured in Zendesk automation rules.\n- Notion output is plain text paragraphs; rich formatting (tables, checkboxes) is not used.\n",[2691,2692,2693,2694,2695],{"_59":402,"_61":403,"_397":24},{"_59":2696,"_61":2697,"_397":242},{"_59":400,"_61":401,"_397":24},{"_59":800,"_61":801,"_397":24},{"_59":395,"_61":396,"_397":242},"zoom","Zoom","Accept a meeting ID via webhook POST, chat, or Slack — extracts it from natural language automatical","Fetch a Zoom meeting transcript via OAuth2","Extract action items from the call, prepended with the account name","Create Zendesk support tickets for issues identified in the call","Create a Notion page with the call summary and action items","Return a clear summary with action items, ticket links, and Notion page link","Report a friendly error when the transcript is still processing","customer-call-follow-up-agent","sales","customer-success","follow-up",{"_465":2739,"_450":2740},{"_450":2734},{"_2184":2714,"_821":2715,"_431":2716,"_433":2717,"_435":2718,"_437":2719,"_2720":2721,"_2722":2723,"_2724":2725},{"_402":2713},{"_426":402},{"_439":440,"_379":2733,"_442":443,"_61":2184,"_445":24},{"_439":440,"_379":2732,"_442":443,"_61":821},{"_439":440,"_379":2731,"_442":443,"_61":431,"_445":24},{"_439":440,"_379":2730,"_442":443,"_61":433,"_445":24},{"_439":440,"_379":444,"_442":443,"_61":435,"_445":24},{"_439":440,"_379":2729,"_442":443,"_61":437},"ZOOM_CLIENT_ID",{"_439":440,"_379":2728,"_442":443,"_61":2720,"_445":24},"ZOOM_CLIENT_SECRET",{"_439":440,"_379":2727,"_442":443,"_61":2722,"_445":24},"ZOOM_REFRESH_TOKEN",{"_439":440,"_379":2726,"_442":443,"_61":2724,"_445":24},"OAuth2 refresh token from Zoom authorization","Zoom marketplace app client secret","Zoom marketplace app client ID","Zendesk subdomain (the part before .zendesk.com, e.g. mycompany)","Zendesk agent email for API auth","Bearer token to authenticate incoming webhook requests","ID of the Notion page that holds all call summary pages","Notion internal integration secret",{"_452":2735,"_454":2736},{"_463":464},{"_456":2737,"_458":2738},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/customer-call-followup:fcbe4509",{"_452":24,"_454":24},"#c46c72","#f13f4c","cruise-line","https://assets.astropods.ai/avatars/agents/astropods/cruise-line.jpg",{"_468":2801,"_470":2802,"_472":2803,"_474":475,"_476":2804,"_478":2805,"_480":710,"_482":2806,"_484":2807},[2752],{"_366":303,"_367":1847},[2749,2750],{"_61":167,"_347":45},{"_61":167,"_347":2751},"chrispattypm",{"_369":2753,"_371":2754,"_373":2755,"_375":2756,"_377":2757},"e045ca73",{"_416":2773,"_958":2774,"_862":2775,"_422":2776,"_371":584},"---\ndescription: \"AI-powered pull request analysis agent — reviews code for correctness, security, and quality with actionable findings, one-click commenting, and fix prompts for Claude Code.\"\ntags:\n - \"code-review\"\n - \"github\"\n - \"pull-requests\"\n - \"security\"\nauthors: []\ncapabilities:\n - \"pr-analysis\"\n - \"code-exploration\"\n - \"chat\"\nintegrations:\n - \"anthropic\"\n - \"github\"\n---\n\n

\n $\"Cruise$ \n

\n\n

Cruise Line

\n $\"Cruise$ \n

\n\n

Cruise Line

\n\nCruise Line analyzes GitHub pull requests for correctness, security, maintainability, performance, and code quality. It produces structured findings with severity levels, concrete fix suggestions, and one-click actions to post comments or copy prompts for Claude Code.\n\n## Setup\n\nAfter deploying, visit your agent's URL to start the setup wizard.\n\n### 1. Connect to GitHub\n\nClick **Connect to GitHub** to create a GitHub App via the manifest flow. You'll be redirected to GitHub to authorize the app. For GitHub Enterprise, check the \"I'm using GitHub Enterprise\" box and enter your instance URL first.\n\nTo switch to a different GitHub connection later, return to `/settings` and click **Disconnect and reconnect**.\n\n### 2. Install on repositories\n\nAfter connecting, click **Install on repos** to choose which repositories Cruise Line can access. You can change this at any time from your GitHub App settings.\n\n## Usage\n\n### PR analysis\n\nWhen you open a pull request on a connected repository, Cruise Line posts a comment with a link. Clicking it opens the analysis page, which automatically starts reviewing the PR.\n\nThe analysis produces:\n\n- **Verdict** — Approve, request changes, or needs discussion, with a rationale\n- **Findings** — Discrete issues ordered by severity (critical, high, medium, low, info), each with a category (correctness, security, maintainability, performance, style)\n- **Suggested changes** — Before/after diffs you can post as GitHub review comments in one click\n- **Fix prompts** — Copy a prompt to paste into Claude Code that gives it all the context to fix the issue\n\n### Chat\n\nSwitch to the **Chat** tab to ask follow-up questions. Claude has full access to the codebase and can use the same interactive directives — suggestions, callouts, and structured findings — in its responses.\n\n### Review rules\n\nOpen **Review rules** from the three-dot menu to configure per-repo rules. Rules are numbered and guide the analysis — for example:\n\n- *Rule #1:* All API endpoints must validate input with Zod\n- *Rule #2:* Never use string interpolation in SQL queries\n- *Rule #3:* React components must not call hooks conditionally\n\nThe agent references rules by number when they're relevant (e.g., \"This violates Rule #2\"). Hovering a rule reference in the analysis shows the full rule text. You can also save findings as new rules directly from the analysis view.\n\n### Commenting\n\nClick the **Comment** button on any code block or suggestion to open the file viewer with the comment input pre-filled. Edit if needed, then submit — it posts directly as a GitHub review comment on the PR.\n\n## Configuration\n\n| Setting | Default | Description |\n|---|---|---|\n| Claude Model | `claude-opus-4-8` | Model used for analysis. Opus is most thorough, Haiku is fastest. |\n| Max Concurrent Jobs | `3` | How many PRs can be analyzed simultaneously. |\n",[2764,2765],{"_59":568,"_61":569,"_397":24},{"_59":408,"_61":1337,"_397":24},"pr-analysis","code-exploration","chat","https://github.com/astropods/cruise-line","code-review","pull-requests","security",{"_465":2788,"_420":2789,"_450":2790},{"_2778":2779,"_2780":2781},{"_351":1489},{"_568":2777},{"_426":568},"chat-sandbox",{"_1010":2782},"cruise-db",{"_426":965},{"_2783":2784,"_465":2785,"_463":464,"_1952":1953},"environment",{"_2786":2787},"registry.astropods.ai/cd210ffd-25bf-4acb-9158-da91e59a883a/cruise-line-knowledge-chat-sandbox:e045ca73","OTEL_EXPORTER_OTLP_ENDPOINT","http://cruise-line-collector:4318","registry.astropods.ai/cd210ffd-25bf-4acb-9158-da91e59a883a/cruise-line:e045ca73",[2791,2792],{"_452":24},{"_439":440,"_642":1259,"_379":2796,"_442":990,"_61":2797,"_991":2798},{"_439":671,"_642":2793,"_379":2794,"_61":2795},"3","Maximum concurrent analysis jobs","MAX_CONCURRENT_JOBS","Claude model used for PR analysis. Opus is the most thorough, Haiku is the fastest.","CLAUDE_MODEL",[1259,2799,2800,647,646],"claude-opus-4-7","claude-opus-4-6","#2c5e76","#9edcfa","#076b9b","#2d6986","#0b1b22","#99c7e6","#85c3e0","create-jira-tickets","https://assets.astropods.ai/avatars/agents/astro-ai/create-jira-tickets.jpg",{"_468":2864,"_470":2865,"_472":2866,"_474":475,"_476":2867,"_478":1587,"_480":1588,"_482":2868,"_484":2869},[2815],{"_366":303,"_367":303},[2814],{"_61":192,"_347":41},{"_369":2816,"_371":2817,"_373":2818,"_375":2819,"_377":2820},"0567a743",{"_416":2838,"_418":2839,"_420":2840,"_422":2841,"_371":424},"---\ndescription: \"Turns a problem description or a Slack thread URL into a Jira ticket in seconds using GPT-4o mini.\"\ntags:\n - jira\n - slack\n - productivity\n - project-management\n - openai\n - issue-tracking\ncapabilities:\n - \"Create a Jira task from a plain-text problem description\"\n - \"Fetch a Slack thread by URL and create a ticket from the conversation content\"\n - \"Generate a concise title and detailed description using GPT-4o mini\"\n - \"Return the direct Atlassian URL to the created ticket\"\nrepository:\n type: github\n url: https://github.com/astropods/agents\n directory: slack-jira-agent\nintegrations:\n - OpenAI\n - Slack\n - Jira\n---\n\n# Slack to Jira Agent\n\nThe bug is sitting in a Slack thread and nobody has filed a ticket yet. Slack to Jira Agent takes a problem description — typed directly, pasted from a conversation, or provided as a Slack thread URL — and turns it into a properly formatted Jira task in seconds. Works from the web playground or directly in Slack via the Slack adapter.\n\n## Usage\n\nSend a message in one of two ways:\n\n**Plain-text description** — describe the problem directly:\n- *\"Login button broken on mobile Safari — users get a 403 after OAuth redirect\"*\n- *\"Checkout flow crashes when applying a discount code on the order summary page\"*\n\n**Slack thread URL** — paste a thread link to create a ticket from the actual conversation (requires `SLACK_BOT_TOKEN`):\n- `https://myworkspace.slack.com/archives/C01234/p1234567890123456`\n\n## What happens\n\n1. If a Slack thread URL is detected, the agent fetches the thread messages via the Slack API\n2. GPT-4o mini generates a concise ticket title (max 100 chars) and a detailed description\n3. The ticket is created in Jira as a `Task` under the configured project\n4. The agent returns the direct link to the created ticket\n\n## Slack integration\n\nThe agent supports both the **web** and **Slack** adapters. Enable the Slack adapter at deploy time and your team can create Jira tickets directly from any Slack channel by messaging the bot — no copy-pasting into a web form.\n\n## Environment variables\n\n| Variable | Description |\n|----------|-------------|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `JIRA_API_KEY` | Jira API token — from Atlassian account security settings |\n| `JIRA_USERNAME` | Jira account email address |\n| `JIRA_SUBDOMAIN` | Subdomain for your Atlassian instance (e.g. `mycompany`) — alphanumeric and hyphens only |\n| `JIRA_PROJECT_ID` | Jira project key (e.g. `PROJ`) |\n| `SLACK_BOT_TOKEN` | *(Optional)* Slack Bot token (`xoxb-…`) — required only when pasting a Slack thread URL; needs `channels:history` scope |\n\n## Limitations\n\n- Creates tickets as `Task` type only; issue type is not configurable at runtime.\n- Jira description is plain text (ADF paragraph); rich formatting is not preserved.\n- Ticket quality depends on the detail in the description — the more context provided, the better the generated content.\n",{"_379":2821,"_381":2822,"_383":2823,"_385":2824,"_387":2825,"_389":2826},"2026-09-28T19:34:37Z","Turns a problem description or a Slack thread URL into a Jira ticket in seconds using GPT-4o mini.",[1338,458,2835,2836,402,2837],{"_62":408,"_409":410,"_411":2834},[2830,2831,2832,2833],"\n# Slack to Jira Agent\n\nThe bug is sitting in a Slack thread and nobody has filed a ticket yet. Slack to Jira Agent takes a problem description — typed directly, pasted from a conversation, or provided as a Slack thread URL — and turns it into a properly formatted Jira task in seconds. Works from the web playground or directly in Slack via the Slack adapter.\n\n## Usage\n\nSend a message in one of two ways:\n\n**Plain-text description** — describe the problem directly:\n- *\"Login button broken on mobile Safari — users get a 403 after OAuth redirect\"*\n- *\"Checkout flow crashes when applying a discount code on the order summary page\"*\n\n**Slack thread URL** — paste a thread link to create a ticket from the actual conversation (requires `SLACK_BOT_TOKEN`):\n- `https://myworkspace.slack.com/archives/C01234/p1234567890123456`\n\n## What happens\n\n1. If a Slack thread URL is detected, the agent fetches the thread messages via the Slack API\n2. GPT-4o mini generates a concise ticket title (max 100 chars) and a detailed description\n3. The ticket is created in Jira as a `Task` under the configured project\n4. The agent returns the direct link to the created ticket\n\n## Slack integration\n\nThe agent supports both the **web** and **Slack** adapters. Enable the Slack adapter at deploy time and your team can create Jira tickets directly from any Slack channel by messaging the bot — no copy-pasting into a web form.\n\n## Environment variables\n\n| Variable | Description |\n|----------|-------------|\n| `OPENAI_API_KEY` | Auto-injected by Astropods |\n| `JIRA_API_KEY` | Jira API token — from Atlassian account security settings |\n| `JIRA_USERNAME` | Jira account email address |\n| `JIRA_SUBDOMAIN` | Subdomain for your Atlassian instance (e.g. `mycompany`) — alphanumeric and hyphens only |\n| `JIRA_PROJECT_ID` | Jira project key (e.g. `PROJ`) |\n| `SLACK_BOT_TOKEN` | *(Optional)* Slack Bot token (`xoxb-…`) — required only when pasting a Slack thread URL; needs `channels:history` scope |\n\n## Limitations\n\n- Creates tickets as `Task` type only; issue type is not configurable at runtime.\n- Jira description is plain text (ADF paragraph); rich formatting is not preserved.\n- Ticket quality depends on the detail in the description — the more context provided, the better the generated content.\n",[2827,2828,2829],{"_59":402,"_61":403,"_397":24},{"_59":458,"_61":1083,"_397":24},{"_59":1338,"_61":1339,"_397":24},"Create a Jira task from a plain-text problem description","Fetch a Slack thread by URL and create a ticket from the conversation content","Generate a concise title and detailed description using GPT-4o mini","Return the direct Atlassian URL to the created ticket","slack-jira-agent","productivity","project-management","issue-tracking",{"_465":2862,"_450":2863},{"_450":2858},{"_2843":2844,"_2845":2846,"_2847":2848,"_2849":2850,"_2851":2852},{"_402":2842},{"_426":402},"JIRA_API_KEY",{"_439":440,"_379":2857,"_442":443,"_61":2843,"_445":24},"JIRA_PROJECT_ID",{"_439":440,"_379":2856,"_442":443,"_61":2845},"JIRA_SUBDOMAIN",{"_439":440,"_379":2855,"_442":443,"_61":2847},"JIRA_USERNAME",{"_439":440,"_379":2854,"_442":443,"_61":2849,"_445":24},"SLACK_BOT_TOKEN",{"_439":440,"_379":2853,"_442":443,"_61":2851,"_445":24},"Slack Bot OAuth token (xoxb-...) — required for fetching Slack thread content when a thread URL is provided","Jira account email address","Jira subdomain (e.g. 'mycompany' for mycompany.atlassian.net)","Jira project key (e.g. 'PROJ')","Jira API token for authentication",{"_454":2859},{"_456":2860,"_458":2861},[462,458],{"_460":24,"_461":24},"registry.astropods.ai/be8e0afc-50d2-44f7-8197-8bc4d1a11833/create-jira-tickets:0567a743",{"_454":24},"#873062","#fa9ed3","#b20568","#862d60","#e29c9e","#e085b9","create-enterprise-page","https://assets.astropods.ai/avatars/agents/postman/create-enterprise-page.jpg?v=1787754977",{"_468":834,"_470":835,"_472":836,"_474":475,"_476":837,"_478":838,"_480":839,"_482":840,"_484":841},[2880],{"_366":2879,"_367":475},[2876],{"_61":2877,"_347":2878},"Roni Lockwood","ronilockwood",50,{"_369":2881,"_371":2882,"_373":2883,"_375":2884,"_377":2885},"981f961d",{"_416":2897,"_418":2898,"_422":2899,"_371":584},"---\ndescription: \"Scaffolds a /for// enterprise pre-sale landing page in postman-eng/marketing-site-www-next-app and opens a PR. Accepts pasted Postman Presales Account Template content or a public Google Doc URL.\"\ntags:\n - enterprise\n - marketing\n - presales\nauthors: []\ncapabilities:\n - create-enterprise-presale-page\nintegrations:\n - \"anthropic\"\n - \"github\"\n---\n\n# create-enterprise-page\n\nAstro Agent that turns a filled-in Postman Presales Account Template (pasted text or public Google Doc URL) into a `page-data/for/.json` file in `postman-eng/marketing-site-www-next-app`, opening a PR against `develop`.\n\nThe full spec is in [`agent/skills/create-enterprise-presale-page/SKILL.md`](agent/skills/create-enterprise-presale-page/SKILL.md). It is loaded as the agent's instructions at startup.\n\n## Required env vars\n\n| Var | Purpose |\n|-----|---------|\n| `ANTHROPIC_API_KEY` | Auto-injected by the anthropic model integration. |\n| `GITHUB_TOKEN` | Token with Contents + Pull requests write on `postman-eng/marketing-site-www-next-app`. Used by the `create_pr` tool. |\n\nSee [README.md → GitHub token setup](README.md#github-token-setup) for setup steps.\n",{"_379":2886,"_381":2887,"_385":2888,"_387":2889,"_389":2890},"2026-09-15T17:04:42Z","Scaffolds a /for// enterprise pre-sale landing page in postman-eng/marketing-site-www-next-app and opens a PR. Accepts pasted Postman Presales Account Template content or a public Google Doc ",[2894,2895,2896],[2893],"\n# create-enterprise-page\n\nAstro Agent that turns a filled-in Postman Presales Account Template (pasted text or public Google Doc URL) into a `page-data/for/.json` file in `postman-eng/marketing-site-www-next-app`, opening a PR against `develop`.\n\nThe full spec is in [`agent/skills/create-enterprise-presale-page/SKILL.md`](agent/skills/create-enterprise-presale-page/SKILL.md). It is loaded as the agent's instructions at startup.\n\n## Required env vars\n\n| Var | Purpose |\n|-----|---------|\n| `ANTHROPIC_API_KEY` | Auto-injected by the anthropic model integration. |\n| `GITHUB_TOKEN` | Token with Contents + Pull requests write on `postman-eng/marketing-site-www-next-app`. Used by the `create_pr` tool. |\n\nSee [README.md → GitHub token setup](README.md#github-token-setup) for setup steps.\n",[2891,2892],{"_59":568,"_61":569,"_397":24},{"_59":408,"_61":1337,"_397":24},"create-enterprise-presale-page","enterprise","marketing","presales",{"_465":2904,"_420":2905,"_450":2906},{"_450":2901},{"_568":2900},{"_426":568},{"_454":2902},{"_456":2903},[462],"registry.astropods.ai/postman/create-enterprise-page:981f961d",[2907,2908],{"_452":24,"_454":24},{"_439":440,"_379":2911,"_61":2912},{"_439":440,"_379":2909,"_61":2910,"_445":24},"GitHub App private key in PEM format. Literal \\n sequences are auto-converted to newlines.","GITHUB_APP_PRIVATE_KEY","GitHub App numeric ID (found on the App settings page).","GITHUB_APP_ID","create-customer-story","https://assets.astropods.ai/avatars/agents/postman/create-customer-story.jpg?v=1786127106",{"_468":2950,"_470":2951,"_472":2952,"_474":475,"_476":2953,"_478":2954,"_480":539,"_482":2955,"_484":2956},[2921],{"_366":2920,"_367":616},[2919],{"_61":2877,"_347":2878},108,{"_369":2922,"_371":2923,"_373":2924,"_375":2925,"_377":2926},"046cd78b",{"_416":2936,"_418":2937,"_422":2938,"_371":584},"---\ndescription: \"Scaffolds a /customer-stories// case study page in postman-eng/marketing-site-www-next-app and opens a PR. Accepts pasted/attached case study content or a public Google Doc URL, and updates the page JSON, the filter card, and the translations entry in one commit.\"\ntags:\n - customer-stories\n - case-study\n - marketing\nauthors: []\ncapabilities:\n - create-customer-story\nintegrations:\n - \"anthropic\"\n - \"github\"\n---\n\n# create-customer-story\n\nAstro Agent that turns case study content (pasted text, an attached doc, or a public Google Doc URL) into a `/customer-stories//` page in `postman-eng/marketing-site-www-next-app`, opening a PR against `develop`.\n\nUnlike the `create-enterprise-page` agent (which writes a single JSON file), a customer story needs **three coordinated edits**, all committed together by the `create_pr` tool:\n\n1. Create `page-data/customer-stories/.json`\n2. Insert a filter card at the top of `src/app/components/filters/customer-stories-data.jsx`\n3. Insert a `customerStories.cards.` entry into `translations/en/global.json`\n\n`page-urls/urls.json` is intentionally **not** committed — the build pipeline regenerates it (`npm run pages`) before `next build`.\n\nThe full spec is in [`agent/skills/create-customer-story/SKILL.md`](agent/skills/create-customer-story/SKILL.md). It is loaded as the agent's instructions at startup.\n\n## Required env vars\n\n| Var | Purpose |\n|-----|---------|\n| `ANTHROPIC_API_KEY` | Auto-injected by the anthropic model integration. |\n| `GITHUB_TOKEN` | Token with Contents + Pull requests write on `postman-eng/marketing-site-www-next-app`. Used by the `create_pr` tool. |\n\nSee [README.md → GitHub token setup](README.md#github-token-setup) for setup steps.\n",{"_379":2927,"_381":2928,"_385":2929,"_387":2930,"_389":2931},"2026-09-28T16:01:27Z","Scaffolds a /customer-stories// case study page in postman-eng/marketing-site-www-next-app and opens a PR. Accepts pasted/attached case study content or a public Google Doc URL, and updates the ",[2934,2935,2895],[2913],"\n# create-customer-story\n\nAstro Agent that turns case study content (pasted text, an attached doc, or a public Google Doc URL) into a `/customer-stories//` page in `postman-eng/marketing-site-www-next-app`, opening a PR against `develop`.\n\nUnlike the `create-enterprise-page` agent (which writes a single JSON file), a customer story needs **three coordinated edits**, all committed together by the `create_pr` tool:\n\n1. Create `page-data/customer-stories/.json`\n2. Insert a filter card at the top of `src/app/components/filters/customer-stories-data.jsx`\n3. Insert a `customerStories.cards.` entry into `translations/en/global.json`\n\n`page-urls/urls.json` is intentionally **not** committed — the build pipeline regenerates it (`npm run pages`) before `next build`.\n\nThe full spec is in [`agent/skills/create-customer-story/SKILL.md`](agent/skills/create-customer-story/SKILL.md). It is loaded as the agent's instructions at startup.\n\n## Required env vars\n\n| Var | Purpose |\n|-----|---------|\n| `ANTHROPIC_API_KEY` | Auto-injected by the anthropic model integration. |\n| `GITHUB_TOKEN` | Token with Contents + Pull requests write on `postman-eng/marketing-site-www-next-app`. Used by the `create_pr` tool. |\n\nSee [README.md → GitHub token setup](README.md#github-token-setup) for setup steps.\n",[2932,2933],{"_59":568,"_61":569,"_397":24},{"_59":408,"_61":1337,"_397":24},"customer-stories","case-study",{"_465":2943,"_420":2944,"_450":2945},{"_450":2940},{"_568":2939},{"_426":568},{"_454":2941},{"_456":2942},[462],"registry.astropods.ai/postman/create-customer-story:046cd78b",[2946,2947],{"_452":24,"_454":24},{"_439":440,"_379":2949,"_61":2912},{"_439":440,"_379":2948,"_61":2910,"_445":24},"PEM private key for the GitHub App. The deploy UI strips whitespace; the runtime normalizer rewraps the base64 body.","Numeric GitHub App id. The App is installed on postman-eng/marketing-site-www-next-app with Contents R/W + Pull requests R/W.","#773daa","#cf9efa","#7a07e0","#5c2d86","#180b22","#c1a6d9","#b585e0","coworker","https://assets.astropods.ai/avatars/agents/simon/coworker.jpg",{"_468":3044,"_470":3045,"_472":3046,"_474":475,"_476":773,"_478":774,"_480":710,"_482":775,"_484":776},[2964],{"_366":493,"_367":1048},[2963],{"_61":31,"_347":31},{"_369":2965,"_371":2966,"_373":2967,"_375":2968,"_377":2969},"f144a06d",{"_416":3003,"_418":3004,"_420":3005,"_958":3006,"_422":3007,"_371":1301},"---\ndescription: \"Fork of Array-Ventures/coworker — AI team member that handles tasks, answers questions, manages workflows, and builds internal tools via chat.\"\ntags: [productivity, workspace, coding, google-workspace, mcp, scheduling, whatsapp]\nauthors:\n - name: Simon Guerrier\n account: simwar\ncapabilities:\n - \"Answers questions and drafts content from conversation context\"\n - \"Builds and maintains internal dashboards and tools with git version control\"\n - \"Reads and sends Gmail, manages Calendar events and Drive files\"\n - \"Installs and manages MCP servers via built-in registry\"\n - \"Runs recurring AI workflows on a cron schedule\"\n - \"Interacts with users over WhatsApp\"\n - \"Uploads, manages, and references files in conversations\"\n - \"Switches between AI providers and models at runtime\"\n - \"Installs community skills from ClawHub and skills.sh\"\n - \"Communicates with other agents via the A2A protocol\"\nrepository: github:Simwar/coworker\nintegrations:\n - Anthropic\n - OpenAI\n - Google Workspace\n - WhatsApp\n - GitHub\n - Slack\n---\n\n## Overview\n\nCoworker is an open-source AI agent that acts as a team member rather than a tool. It remembers past conversations, tracks tasks across sessions, manages your Google Workspace, and can build internal dashboards — all through natural language chat. This is a fork of the original [coworker](https://github.com/Array-Ventures/coworker) by Array Ventures.\n\nBuilt on [Mastra](https://mastra.ai) and designed for team deployment on the Astro platform.\n\n## Usage\n\nChat naturally. Coworker picks the right tools based on context:\n\n- **Ask questions** — it searches memory for relevant past context before answering\n- **Request tasks** — it tracks multi-step work with a visible task list and checks in when done\n- **Google Workspace** — \"summarise my unread emails\", \"schedule a meeting Thursday 3pm\", \"share the Q2 doc with the team\"\n- **Build tools** — \"create a dashboard showing our open GitHub issues\" — it scaffolds, commits, and maintains the app\n- **Schedule work** — \"every Monday morning, summarise last week's activity and post to Slack\"\n- **Install skills** — \"install the Jira skill from ClawHub\" to extend what it can do\n\n## Inputs\n\n| Input | Required | Description |\n|-------|----------|-------------|\n| `MODEL` | No | Model to use, e.g. `anthropic/claude-sonnet-4-5` or `openai/gpt-4o` |\n| `ANTHROPIC_API_KEY` | If using Anthropic | Anthropic API key |\n| `OPENAI_API_KEY` | If using OpenAI | OpenAI API key |\n| `GOG_GOOGLE_CLIENT_ID` | For Google Workspace | OAuth client ID from Google Cloud Console |\n| `GOG_GOOGLE_CLIENT_SECRET` | For Google Workspace | OAuth client secret |\n| `GOG_KEYRING_PASSWORD` | Recommended | Encrypts stored Google OAuth tokens at rest |\n\n## Limitations\n\n- Google Workspace requires a Google Cloud OAuth app — see README for setup steps\n- WhatsApp sessions are not persisted across redeployments (re-scan QR on restart)\n- Scheduled tasks run in-process — they pause if the agent container restarts\n- Local LLM providers (LM Studio, Ollama) require the agent to share a network with the provider\n",{"_379":2970,"_381":2971,"_559":2972,"_383":2973,"_385":2974,"_387":2975,"_389":2976},"2026-05-12T10:01:30Z","Fork of Array-Ventures/coworker — AI team member that handles tasks, answers questions, manages workflows, and builds internal tools via chat.",[2835,2999,3000,3001,1693,3002,2983],[2998],{"_62":574,"_409":2997},[2987,2988,2989,2990,2991,2992,2993,2994,2995,2996],"\n## Overview\n\nCoworker is an open-source AI agent that acts as a team member rather than a tool. It remembers past conversations, tracks tasks across sessions, manages your Google Workspace, and can build internal dashboards — all through natural language chat. This is a fork of the original [coworker](https://github.com/Array-Ventures/coworker) by Array Ventures.\n\nBuilt on [Mastra](https://mastra.ai) and designed for team deployment on the Astro platform.\n\n## Usage\n\nChat naturally. Coworker picks the right tools based on context:\n\n- **Ask questions** — it searches memory for relevant past context before answering\n- **Request tasks** — it tracks multi-step work with a visible task list and checks in when done\n- **Google Workspace** — \"summarise my unread emails\", \"schedule a meeting Thursday 3pm\", \"share the Q2 doc with the team\"\n- **Build tools** — \"create a dashboard showing our open GitHub issues\" — it scaffolds, commits, and maintains the app\n- **Schedule work** — \"every Monday morning, summarise last week's activity and post to Slack\"\n- **Install skills** — \"install the Jira skill from ClawHub\" to extend what it can do\n\n## Inputs\n\n| Input | Required | Description |\n|-------|----------|-------------|\n| `MODEL` | No | Model to use, e.g. `anthropic/claude-sonnet-4-5` or `openai/gpt-4o` |\n| `ANTHROPIC_API_KEY` | If using Anthropic | Anthropic API key |\n| `OPENAI_API_KEY` | If using OpenAI | OpenAI API key |\n| `GOG_GOOGLE_CLIENT_ID` | For Google Workspace | OAuth client ID from Google Cloud Console |\n| `GOG_GOOGLE_CLIENT_SECRET` | For Google Workspace | OAuth client secret |\n| `GOG_KEYRING_PASSWORD` | Recommended | Encrypts stored Google OAuth tokens at rest |\n\n## Limitations\n\n- Google Workspace requires a Google Cloud OAuth app — see README for setup steps\n- WhatsApp sessions are not persisted across redeployments (re-scan QR on restart)\n- Scheduled tasks run in-process — they pause if the agent container restarts\n- Local LLM providers (LM Studio, Ollama) require the agent to share a network with the provider\n",[2977,2978,2979,2980,2981,2982],{"_59":568,"_61":569,"_397":24},{"_59":402,"_61":403,"_397":24},{"_59":2985,"_61":2986,"_397":242},{"_59":2983,"_61":2984,"_397":242},{"_59":408,"_61":1337,"_397":24},{"_59":458,"_61":1083,"_397":24},"whatsapp","WhatsApp","google workspace","Google Workspace","Answers questions and drafts content from conversation context","Builds and maintains internal dashboards and tools with git version control","Reads and sends Gmail, manages Calendar events and Drive files","Installs and manages MCP servers via built-in registry","Runs recurring AI workflows on a cron schedule","Interacts with users over WhatsApp","Uploads, manages, and references files in conversations","Switches between AI providers and models at runtime","Installs community skills from ClawHub and skills.sh","Communicates with other agents via the A2A protocol","https://github.com/Simwar/coworker",{"_61":908,"_347":909},"workspace","coding","google-workspace","scheduling",{"_465":3042,"_450":3043},{"_450":3039},{"_3027":3028,"_3029":3030,"_3031":3032,"_1257":3033},{"_3009":3010},{"_568":3008},{"_426":568},"pg",{"_1010":3011,"_420":3012,"_1194":24},{"_465":3024,"_463":3025,"_1952":3026},[3013,3014,3015,3016,3017],{"_439":440,"_642":2957,"_61":1960},{"_439":440,"_642":965,"_61":3023},{"_439":440,"_642":965,"_61":3022},{"_439":440,"_642":3020,"_61":3021},{"_439":440,"_642":3018,"_61":3019},"/var/lib/postgresql/data/pgdata","PGDATA","trust","POSTGRES_HOST_AUTH_METHOD","POSTGRES_PASSWORD","POSTGRES_USER","pgvector/pgvector:pg17",5432,"/var/lib/postgresql/data","GOG_GOOGLE_CLIENT_ID",{"_439":440,"_379":3038,"_61":3027,"_984":24},"GOG_GOOGLE_CLIENT_SECRET",{"_439":440,"_379":3037,"_61":3029,"_984":24,"_445":24},"GOG_KEYRING_PASSWORD",{"_439":440,"_379":3036,"_61":3031,"_984":24,"_445":24},{"_439":440,"_642":3034,"_379":3035,"_61":1257},"anthropic/claude-sonnet-4-5","Model to use (e.g. anthropic/claude-sonnet-4-5, openai/gpt-4o)","Password used to encrypt stored Google OAuth tokens","Google OAuth client secret for Workspace integration","Google OAuth client ID for Workspace integration",{"_454":3040},{"_456":3041},[462],"registry.astropods.ai/simon/coworker:f144a06d",{"_454":24},"#aec5e3","#9ec6fa","#94c2fd","brainstorming","registry.astropods.ai/bdwanless","https://assets.astropods.ai/avatars/agents/bdwanless/brainstorming.jpg",{"_468":3075,"_470":3076,"_472":3077,"_474":475,"_476":2396,"_478":538,"_480":539,"_482":2397,"_484":2398},[3055],{"_366":493,"_367":475},[3054],{"_61":257,"_347":29},{"_369":3056,"_371":3057,"_373":3058,"_375":3059,"_377":3060},"f43a2e0b",{"_416":3067,"_418":3068,"_422":3069,"_371":584},"---\nname: brainstorming\ndescription: \"You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.\"\ntags: []\nauthors: []\ncapabilities: []\nintegrations:\n - \"openai\"\nrepository:\n type: git\n url: \"https://github.com/newsledbans/astro-agent-blueprints\"\n directory: brainstorming\n---\n\n# Brainstorming Ideas Into Designs\n\n## Overview\n\nHelp turn ideas into fully formed designs and specs through natural collaborative dialogue.\n\nStart by understanding the current project context, then ask questions one at a time to refine the idea. Once you understand what you're building, present the design in small sections (200-300 words), checking after each section whether it looks right so far.\n\n## The Process\n\n**Understanding the idea:**\n- Check out the current project state first (files, docs, recent commits)\n- Ask questions one at a time to refine the idea\n- Prefer multiple choice questions when possible, but open-ended is fine too\n- Only one question per message - if a topic needs more exploration, break it into multiple questions\n- Focus on understanding: purpose, constraints, success criteria\n\n**Exploring approaches:**\n- Propose 2-3 different approaches with trade-offs\n- Present options conversationally with your recommendation and reasoning\n- Lead with your recommended option and explain why\n\n**Presenting the design:**\n- Once you believe you understand what you're building, present the design\n- Break it into sections of 200-300 words\n- Ask after each section whether it looks right so far\n- Cover: architecture, components, data flow, error handling, testing\n- Be ready to go back and clarify if something doesn't make sense\n\n## After the Design\n\n**Documentation:**\n- Write the validated design to `docs/plans/YYYY-MM-DD--design.md`\n- Use elements-of-style:writing-clearly-and-concisely skill if available\n- Commit the design document to git\n\n**Implementation (if continuing):**\n- Ask: \"Ready to set up for implementation?\"\n- Use superpowers:using-git-worktrees to create isolated workspace\n- Use superpowers:writing-plans to create detailed implementation plan\n\n## Key Principles\n\n- **One question at a time** - Don't overwhelm with multiple questions\n- **Multiple choice preferred** - Easier to answer than open-ended when possible\n- **YAGNI ruthlessly** - Remove unnecessary features from all designs\n- **Explore alternatives** - Always propose 2-3 approaches before settling\n- **Incremental validation** - Present design in sections, validate each\n- **Be flexible** - Go back and clarify when something doesn't make sense\n",{"_379":3061,"_383":3062,"_387":3063,"_389":3064},"2026-06-05T18:20:30Z","You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.",{"_62":574,"_409":3066,"_411":3047},"\n# Brainstorming Ideas Into Designs\n\n## Overview\n\nHelp turn ideas into fully formed designs and specs through natural collaborative dialogue.\n\nStart by understanding the current project context, then ask questions one at a time to refine the idea. Once you understand what you're building, present the design in small sections (200-300 words), checking after each section whether it looks right so far.\n\n## The Process\n\n**Understanding the idea:**\n- Check out the current project state first (files, docs, recent commits)\n- Ask questions one at a time to refine the idea\n- Prefer multiple choice questions when possible, but open-ended is fine too\n- Only one question per message - if a topic needs more exploration, break it into multiple questions\n- Focus on understanding: purpose, constraints, success criteria\n\n**Exploring approaches:**\n- Propose 2-3 different approaches with trade-offs\n- Present options conversationally with your recommendation and reasoning\n- Lead with your recommended option and explain why\n\n**Presenting the design:**\n- Once you believe you understand what you're building, present the design\n- Break it into sections of 200-300 words\n- Ask after each section whether it looks right so far\n- Cover: architecture, components, data flow, error handling, testing\n- Be ready to go back and clarify if something doesn't make sense\n\n## After the Design\n\n**Documentation:**\n- Write the validated design to `docs/plans/YYYY-MM-DD--design.md`\n- Use elements-of-style:writing-clearly-and-concisely skill if available\n- Commit the design document to git\n\n**Implementation (if continuing):**\n- Ask: \"Ready to set up for implementation?\"\n- Use superpowers:using-git-worktrees to create isolated workspace\n- Use superpowers:writing-plans to create detailed implementation plan\n\n## Key Principles\n\n- **One question at a time** - Don't overwhelm with multiple questions\n- **Multiple choice preferred** - Easier to answer than open-ended when possible\n- **YAGNI ruthlessly** - Remove unnecessary features from all designs\n- **Explore alternatives** - Always propose 2-3 approaches before settling\n- **Incremental validation** - Present design in sections, validate each\n- **Be flexible** - Go back and clarify when something doesn't make sense\n",[3065],{"_59":402,"_61":403,"_397":24},"https://github.com/newsledbans/astro-agent-blueprints",{"_465":3074},{"_450":3071},{"_402":3070},{"_426":402},{"_454":3072},{"_456":3073},[462],"registry.astropods.ai/bdwanless/brainstorming:f43a2e0b","#9b72c9","#ca9ff9","#9947f4","ada-your-personal-coding-agent","https://assets.astropods.ai/avatars/agents/astropods/ada-your-personal-coding-agent.jpg?v=1784911936",{"_468":3196,"_470":3197,"_472":3198,"_474":475,"_476":3199,"_478":3200,"_480":539,"_482":3201,"_484":3202},[3087],{"_366":3085,"_367":3086},[3084],{"_61":192,"_347":41},87,18,{"_369":3088,"_371":3089,"_373":3090,"_375":3091,"_377":3092},"bb842f1a",{"_416":3119,"_418":3120,"_420":3121,"_958":3122,"_422":3123,"_3124":3125,"_371":584},"---\ndescription: \"Ada — a coding agent (built on Flue 2.0) you talk to over Slack, web chat, or an AG-UI/SSE frontend.\"\ntags: [ada, flue, ag-ui, sse, slack, github, coding-agent]\ncapabilities:\n - \"Runs a Flue 2.0 agent loop in-process — no CLI child, no stream-json contract\"\n - \"Secure by default: always-on principles, on-demand security playbooks, and enforced diff gates\"\n - \"Holds multi-turn conversations, durable in Postgres across restarts and redeploys\"\n - \"Runs the model's commands in a separate, secret-less sandbox container\"\n - \"Isolates each session in its own writable workspace\"\n - \"Scrubs the sandbox env so tools can't read host secrets (AWS/DB/gateway)\"\n - \"Translates Flue's event stream into AG-UI protocol events, one-to-one\"\n - \"Re-emits AG-UI events over SSE with lossless reconnect (Last-Event-ID replay)\"\n - \"Serves an AG-UI-compatible frontend; external AG-UI clients (e.g. CopilotKit) can consume the stream\"\n - \"Also exposes Flue's native agent routes for @flue/sdk clients\"\n - \"Connects each user's GitHub via per-user OAuth so git/gh act as them\"\n - \"Accepts file attachments from Slack, web chat, and the AG-UI frontend — images are shown to the model directly, other files are staged at uploads/ for its own tools to read; oversized or too-many-file batches are refused with an explanation\"\n - \"Exports OTel traces/metrics via @flue/opentelemetry; content capture off by default\"\nrepository: github:astropods/ada\nintegrations: [GitHub, Slack]\n---\n\n## Overview\n\n**Ada** is a coding agent built on [Flue 2.0](https://flueframework.com), the\nopen TypeScript agent framework. The agent is a function — `Ada()` in\n`agent/agents/ada.ts` — that declares what it needs through hooks: `useModel()`\nfor the LLM, `useSandbox()` for a filesystem and shell, `useInstruction()` for\nguidance, `useTool()` for capabilities. Flue runs the loop in-process.\n\nConverse with Ada over Slack / web chat (the platform messaging sidecar) or any\nAG-UI frontend. Flue's event stream is translated into\n[AG-UI](https://docs.ag-ui.com) protocol events and streamed to the browser over\nServer-Sent Events, so any AG-UI-compatible frontend can render a run — a plain\nchat UI or a richer dashboard — without backend changes.\n\nEach session gets its own writable sandbox workspace as its working directory.\nWhen the sandbox container is deployed, that workspace is also isolated FROM\nOTHER SESSIONS two ways: path resolution rejects a file operation that escapes\nit, and shell commands run under a per-instance uid against a 0700 directory, so\none conversation cannot read another's checkout by naming an absolute path\neither. Command confinement needs the sandbox service to run as root and is\nreported by its `/health`; the network is not isolated. The in-process `local()`\nfallback has neither — it is for local runs, not a multi-tenant boundary.\n\nThe sandbox env is scrubbed: Flue's `local()` allowlist plus the proxy/TLS\nand build-cache vars, and the user's GitHub token. So a command the model runs —\n`bash` → `env` — can't read the container's host secrets (AWS, database, gateway\nkeys). The model credential is no longer forwarded at all, because no child\nprocess needs it.\n\nModel calls route through the Astro AI Gateway (`astro_ai_gateway: true`) via a\ncustom pi-ai provider, so the container needs no personal `ANTHROPIC_API_KEY`.\nSupplying one as an input opts out of the gateway and talks to Anthropic\ndirectly.\n\n> **Why the name?** Named after **Ada Lovelace** — widely regarded as the first\n> computer programmer — a fitting namesake for a coding agent. The Astropods\n> blueprint deploys as `ada-your-personal-coding-agent` because a bare `ada`\n> isn't a valid blueprint name.\n\n## Usage\n\nMessage Ada over Slack / web chat, or open the agent's frontend URL and send a\nprompt (e.g. *\"read config.json and bump the retries\"*). You'll see streaming\nassistant text, tool-call cards with live arguments, tool results, and a final\nresult — as AG-UI events. Keep replying to continue the same session: follow-up\nturns resume the prior context in the same workspace. The built-in UI shows a\nper-session token/cost readout, a session-history switcher, and a settings panel\n(GitHub status + connect/setup actions).\n\nExternal clients can drive the same pipeline over HTTP:\n\n- **`POST /sessions`** `{prompt}` → `{sessionId}` (starts a session)\n- **`POST /sessions/:id/messages`** `{prompt}` → continue that session with a follow-up turn\n- **`GET /sessions/:id/events`** → SSE AG-UI stream; honours `Last-Event-ID` for lossless replay\n\nFlue's own routes are mounted alongside at **`/agents/ada/:id`** for `@flue/sdk`\nclients. Both address the same durable instances, so a session started over one\nis readable over the other — which is why `/agents/ada/*` is admin-gated rather\nthan open, and why every `/sessions/:id` route checks session OWNERSHIP and not\njust whether the caller is authorized. See \"Who may touch which session\" in the\nREADME; set `ADMIN_EMAILS` in any multi-user deployment.\n\nThe native router is **read/abort only**: `POST /agents/ada/:id` (prompt\ndispatch) returns 405. It renders the agent without `SessionRunner`, so the\nper-session context that selects the sandbox was never published and the turn\nexecuted in the control-plane container at `process.cwd()` instead. Start turns\nwith `POST /sessions`.\n\nThree ways to run it:\n\n```bash\nnode agent/index.ts # no build step (type-stripping)\nvite build && node dist/server.mjs # the Flue Node target\nnpx flue run agent/agents/ada.ts --message \"…\" # one turn, terminal only\n```\n\n## GitHub\n\nAda can act on GitHub as the individual user it's talking to — no static token\nbaked into the container.\n\n- **Connect** — a user runs `connect github` in a DM (or clicks **Connect\n GitHub** in the web UI); Ada replies with a link to authorize. Their token is\n stored per-user and injected as `GH_TOKEN` into the sandbox env, so `git` and\n `gh` inside the session act as them. A connected user's render also gains the\n `checkout_repo` tool, which clones with that token server-side — the model\n names `owner/repo` and never handles a credential.\n- **Set up** — an admin (whose verified email is listed in `ADMIN_EMAILS`) opens\n the web UI and clicks **Set up GitHub App** to provision the App through\n GitHub's App-Manifest flow. No manual App registration, no client id/secret to\n copy. The App is provisioned **private** by default (single-tenant); set\n `GITHUB_APP_PUBLIC=true` for multi-tenant so other users can install it on their\n own repos and connect. (Setup is web-only: admin is verified by email, which chat doesn't carry.\n If `ADMIN_EMAILS` is unset, any signed-in user can perform the *initial* setup;\n provisioning then locks.)\n\n## Secure by default\n\nAda is opinionated about how code gets written, in three layers: a short set of\nalways-on principles, seven security playbooks loaded on demand as Flue skills\n(API authorization, safe migrations, injection defence, docs, dependencies,\nsecrets, agent integrity), and enforcement gates that read the real diff at\nevery would-stop point and send the model back when a default was missed.\n\nBefore planning, Ada also reads the repository's own conventions\n(`AGENTS.md`/`AGENT.md`/`CLAUDE.md`/`CONVENTIONS.md`, matched case-insensitively\nin the working directory or a checkout one level below it) and treats them as\noverriding its own defaults — so project rules like \"every PR needs a changelog\nfile\" are followed rather than discovered by CI.\n\nA further layer runs on substantive changes: an independent `code-reviewer`\nsubagent whose lenses are ported from Cruise Line (test quality, error handling,\nAPI contract, concurrency). It reviews with fresh context — an agent re-reading\nits own work has already convinced itself — and its critical/high findings send\nAda back before finishing. Enabled by `ADA_REVIEWER_ENABLED`, and switchable\nmid-conversation by just asking Ada to turn review off.\n\nThe gates block on credentials appearing in a diff and on controls being\nweakened to reach green (a skipped test, a `@ts-ignore`, `--no-verify`), and\nadvise on endpoints changed without tests, stale API specs, migrations without a\nreversal, and dependencies added without a lockfile. Advisory rules fire once\nper session and blocking ones escalate three times then stop, so a gate can\nnever loop.\n\n## Limitations\n\n- Flue's conversation database and the sandbox workspaces live on the container\n filesystem, so **conversation history does not survive a restart or\n redeploy** — a resumed Slack thread re-clones its repo and starts fresh. A\n volume can only attach to a knowledge *container*, and the out-of-process\n sandbox that used to provide one is gone; the durable fix is a\n Postgres-backed Flue `PersistenceAdapter`. GitHub tokens and repo bindings\n are unaffected (those are in Postgres).\n- Locally, with no database, conversations are in SQLite and do not survive a\n restart. Deployed they are in Postgres and do.\n- The AG-UI event log is in memory with no GC yet. (The OAuth CSRF state is no\n longer among these — it is a signed, self-contained token now.)\n- One turn runs at a time per session — an overlapping follow-up gets `409 busy`.\n",{"_379":3093,"_381":3094,"_383":3095,"_385":3096,"_387":3097,"_389":3098},"2026-10-12T15:11:56Z","Ada — a coding agent (built on Flue 2.0) you talk to over Slack, web chat, or an AG-UI/SSE frontend.",[3115,3116,2323,3117,458,408,3118],{"_62":574,"_409":3114},[3101,3102,3103,3104,3105,3106,3107,3108,3109,3110,3111,3112,3113],"\n## Overview\n\n**Ada** is a coding agent built on [Flue 2.0](https://flueframework.com), the\nopen TypeScript agent framework. The agent is a function — `Ada()` in\n`agent/agents/ada.ts` — that declares what it needs through hooks: `useModel()`\nfor the LLM, `useSandbox()` for a filesystem and shell, `useInstruction()` for\nguidance, `useTool()` for capabilities. Flue runs the loop in-process.\n\nConverse with Ada over Slack / web chat (the platform messaging sidecar) or any\nAG-UI frontend. Flue's event stream is translated into\n[AG-UI](https://docs.ag-ui.com) protocol events and streamed to the browser over\nServer-Sent Events, so any AG-UI-compatible frontend can render a run — a plain\nchat UI or a richer dashboard — without backend changes.\n\nEach session gets its own writable sandbox workspace as its working directory.\nWhen the sandbox container is deployed, that workspace is also isolated FROM\nOTHER SESSIONS two ways: path resolution rejects a file operation that escapes\nit, and shell commands run under a per-instance uid against a 0700 directory, so\none conversation cannot read another's checkout by naming an absolute path\neither. Command confinement needs the sandbox service to run as root and is\nreported by its `/health`; the network is not isolated. The in-process `local()`\nfallback has neither — it is for local runs, not a multi-tenant boundary.\n\nThe sandbox env is scrubbed: Flue's `local()` allowlist plus the proxy/TLS\nand build-cache vars, and the user's GitHub token. So a command the model runs —\n`bash` → `env` — can't read the container's host secrets (AWS, database, gateway\nkeys). The model credential is no longer forwarded at all, because no child\nprocess needs it.\n\nModel calls route through the Astro AI Gateway (`astro_ai_gateway: true`) via a\ncustom pi-ai provider, so the container needs no personal `ANTHROPIC_API_KEY`.\nSupplying one as an input opts out of the gateway and talks to Anthropic\ndirectly.\n\n> **Why the name?** Named after **Ada Lovelace** — widely regarded as the first\n> computer programmer — a fitting namesake for a coding agent. The Astropods\n> blueprint deploys as `ada-your-personal-coding-agent` because a bare `ada`\n> isn't a valid blueprint name.\n\n## Usage\n\nMessage Ada over Slack / web chat, or open the agent's frontend URL and send a\nprompt (e.g. *\"read config.json and bump the retries\"*). You'll see streaming\nassistant text, tool-call cards with live arguments, tool results, and a final\nresult — as AG-UI events. Keep replying to continue the same session: follow-up\nturns resume the prior context in the same workspace. The built-in UI shows a\nper-session token/cost readout, a session-history switcher, and a settings panel\n(GitHub status + connect/setup actions).\n\nExternal clients can drive the same pipeline over HTTP:\n\n- **`POST /sessions`** `{prompt}` → `{sessionId}` (starts a session)\n- **`POST /sessions/:id/messages`** `{prompt}` → continue that session with a follow-up turn\n- **`GET /sessions/:id/events`** → SSE AG-UI stream; honours `Last-Event-ID` for lossless replay\n\nFlue's own routes are mounted alongside at **`/agents/ada/:id`** for `@flue/sdk`\nclients. Both address the same durable instances, so a session started over one\nis readable over the other — which is why `/agents/ada/*` is admin-gated rather\nthan open, and why every `/sessions/:id` route checks session OWNERSHIP and not\njust whether the caller is authorized. See \"Who may touch which session\" in the\nREADME; set `ADMIN_EMAILS` in any multi-user deployment.\n\nThe native router is **read/abort only**: `POST /agents/ada/:id` (prompt\ndispatch) returns 405. It renders the agent without `SessionRunner`, so the\nper-session context that selects the sandbox was never published and the turn\nexecuted in the control-plane container at `process.cwd()` instead. Start turns\nwith `POST /sessions`.\n\nThree ways to run it:\n\n```bash\nnode agent/index.ts # no build step (type-stripping)\nvite build && node dist/server.mjs # the Flue Node target\nnpx flue run agent/agents/ada.ts --message \"…\" # one turn, terminal only\n```\n\n## GitHub\n\nAda can act on GitHub as the individual user it's talking to — no static token\nbaked into the container.\n\n- **Connect** — a user runs `connect github` in a DM (or clicks **Connect\n GitHub** in the web UI); Ada replies with a link to authorize. Their token is\n stored per-user and injected as `GH_TOKEN` into the sandbox env, so `git` and\n `gh` inside the session act as them. A connected user's render also gains the\n `checkout_repo` tool, which clones with that token server-side — the model\n names `owner/repo` and never handles a credential.\n- **Set up** — an admin (whose verified email is listed in `ADMIN_EMAILS`) opens\n the web UI and clicks **Set up GitHub App** to provision the App through\n GitHub's App-Manifest flow. No manual App registration, no client id/secret to\n copy. The App is provisioned **private** by default (single-tenant); set\n `GITHUB_APP_PUBLIC=true` for multi-tenant so other users can install it on their\n own repos and connect. (Setup is web-only: admin is verified by email, which chat doesn't carry.\n If `ADMIN_EMAILS` is unset, any signed-in user can perform the *initial* setup;\n provisioning then locks.)\n\n## Secure by default\n\nAda is opinionated about how code gets written, in three layers: a short set of\nalways-on principles, seven security playbooks loaded on demand as Flue skills\n(API authorization, safe migrations, injection defence, docs, dependencies,\nsecrets, agent integrity), and enforcement gates that read the real diff at\nevery would-stop point and send the model back when a default was missed.\n\nBefore planning, Ada also reads the repository's own conventions\n(`AGENTS.md`/`AGENT.md`/`CLAUDE.md`/`CONVENTIONS.md`, matched case-insensitively\nin the working directory or a checkout one level below it) and treats them as\noverriding its own defaults — so project rules like \"every PR needs a changelog\nfile\" are followed rather than discovered by CI.\n\nA further layer runs on substantive changes: an independent `code-reviewer`\nsubagent whose lenses are ported from Cruise Line (test quality, error handling,\nAPI contract, concurrency). It reviews with fresh context — an agent re-reading\nits own work has already convinced itself — and its critical/high findings send\nAda back before finishing. Enabled by `ADA_REVIEWER_ENABLED`, and switchable\nmid-conversation by just asking Ada to turn review off.\n\nThe gates block on credentials appearing in a diff and on controls being\nweakened to reach green (a skipped test, a `@ts-ignore`, `--no-verify`), and\nadvise on endpoints changed without tests, stale API specs, migrations without a\nreversal, and dependencies added without a lockfile. Advisory rules fire once\nper session and blocking ones escalate three times then stop, so a gate can\nnever loop.\n\n## Limitations\n\n- Flue's conversation database and the sandbox workspaces live on the container\n filesystem, so **conversation history does not survive a restart or\n redeploy** — a resumed Slack thread re-clones its repo and starts fresh. A\n volume can only attach to a knowledge *container*, and the out-of-process\n sandbox that used to provide one is gone; the durable fix is a\n Postgres-backed Flue `PersistenceAdapter`. GitHub tokens and repo bindings\n are unaffected (those are in Postgres).\n- Locally, with no database, conversations are in SQLite and do not survive a\n restart. Deployed they are in Postgres and do.\n- The AG-UI event log is in memory with no GC yet. (The OAuth CSRF state is no\n longer among these — it is a signed, self-contained token now.)\n- One turn runs at a time per session — an overlapping follow-up gets `409 busy`.\n",[3099,3100],{"_59":408,"_61":1337,"_397":24},{"_59":458,"_61":1083,"_397":24},"Runs a Flue 2.0 agent loop in-process — no CLI child, no stream-json contract","Secure by default: always-on principles, on-demand security playbooks, and enforced diff gates","Holds multi-turn conversations, durable in Postgres across restarts and redeploys","Runs the model's commands in a separate, secret-less sandbox container","Isolates each session in its own writable workspace","Scrubs the sandbox env so tools can't read host secrets (AWS/DB/gateway)","Translates Flue's event stream into AG-UI protocol events, one-to-one","Re-emits AG-UI events over SSE with lossless reconnect (Last-Event-ID replay)","Serves an AG-UI-compatible frontend; external AG-UI clients (e.g. CopilotKit) can consume the stream","Also exposes Flue's native agent routes for @flue/sdk clients","Connects each user's GitHub via per-user OAuth so git/gh act as them","Accepts file attachments from Slack, web chat, and the AG-UI frontend — images are shown to the mode","Exports OTel traces/metrics via @flue/opentelemetry; content capture off by default","https://github.com/astropods/ada","ada","flue","sse","coding-agent",{"_465":3194,"_450":3195},{"_586":648,"_450":3190},{"_3165":3166,"_3167":3168,"_2587":3169,"_966":3170,"_3171":3172,"_3173":3174,"_3175":3176},{"_3160":3161,"_3124":3162},{"_642":3159},"sandbox",{"_3126":3127,"_3128":3129,"_3130":3131,"_3132":3133},"env",{"_3140":2339,"_3141":1261,"_3142":2339,"_3143":3144,"_3145":2339,"_3146":3147,"_3148":3149,"_3150":2339,"_3151":2339,"_3152":2339,"_3153":3154,"_3155":2339,"_3156":2339,"_3157":2339,"_3158":2339},"node",{"_3136":3137,"_474":3138},"setup",[3134,3135],"toolchain","auto","command -v gh >/dev/null || { curl -fsSL https://github.com/cli/cli/releases/download/v2.101.0/gh_2.101.0_linux_arm64.tar.gz | tar xz -C /tmp && install -m 0755 /tmp/gh_2.101.0_linux_arm64/bin/gh /opt/astropods/bin/gh; }","git config --system credential.\"https://github.com\".helper \"!gh auth git-credential\"","packages",[3139],"24","bun@1.3.9","CARGO_BUILD_JOBS","GIT_TERMINAL_PROMPT","GOMAXPROCS","GOMEMLIMIT","1536MiB","JOBS","MAKEFLAGS","-j1","NODE_OPTIONS","--max-old-space-size=1024","NX_PARALLEL","PYTHONDONTWRITEBYTECODE","TURBO_CONCURRENCY","UV_THREADPOOL_SIZE","2","VITEST_MAX_FORKS","VITEST_MAX_THREADS","VITEST_MIN_THREADS","npm_config_jobs",{"_426":645},"ada-db",{"_426":965},{"_1010":3163},{"_465":3164,"_463":464,"_1952":1953},"registry.astropods.ai/cd210ffd-25bf-4acb-9158-da91e59a883a/ada-your-personal-coding-agent-knowledge-sandbox:bb842f1a","ADA_REVIEWER_ENABLED",{"_439":440,"_642":1658,"_379":3188,"_442":990,"_61":3165,"_984":24,"_991":3189},"ADA_SANDBOX",{"_439":440,"_642":3133,"_379":3184,"_442":990,"_61":3167,"_984":24,"_991":3185},{"_439":440,"_379":3183,"_442":443,"_61":2587,"_984":24},{"_439":440,"_379":3182,"_442":443,"_61":966,"_984":24,"_445":24},"GITHUB_APP_PUBLIC",{"_439":440,"_642":3179,"_379":3180,"_442":990,"_61":3171,"_984":24,"_991":3181},"MAX_CONCURRENT_RUNS",{"_439":671,"_642":2793,"_379":3178,"_442":443,"_61":3173,"_984":24},"PUBLIC_BASE_URL",{"_439":440,"_379":3177,"_442":443,"_61":3175,"_984":24},"Base URL a browser uses to reach this agent, e.g. http://localhost:3200. Leave empty to derive it; required for GitHub connect links sent to chat.","Max concurrent agent runs. Lower to 1–2 to ease container memory pressure.","false","Provision the GitHub App as public (multi-tenant: installable by other users). Default false = private/single-tenant.",[3179,1658],"Optional: use this Anthropic API key directly and bypass the Astro AI Gateway","Optional: comma-separated admin emails allowed to set up the GitHub App (matched against the verified login email). Unset = any signed-in user can do the initial setup.","Where the model's commands run. auto = as injected (container, else in-process); astro = a platform MicroVM per conversation; local = in-process, NOT isolated.",[3133,3186,1010,3187],"astro","local","Run an independent code-reviewer pass on substantive changes. Can be toggled per conversation by asking Ada.",[1658,3179],{"_452":3191,"_454":3192},{"_463":1956},{"_456":3193},[462,458],"registry.astropods.ai/cd210ffd-25bf-4acb-9158-da91e59a883a/ada-your-personal-coding-agent:bb842f1a",{"_452":24,"_454":24},"#d4cddf","#c7b6e2","#d2c4e8","#513280","#150d21","#b9a3db","#ab91d4","status","unauthenticated","canRead","canEdit","canOperate","canManageGitHub","canArchive","canManageAccess","uid","e7f6c8d3-6e28-4bdb-9100-81ad7f6cf2b8",{"_468":1494,"_470":1495,"_472":1496,"_474":475,"_476":1497,"_478":1498,"_480":1274,"_482":775,"_484":1499},[3218,3219,3220,3221,3222],{"_366":1465,"_367":1466},[3217],{"_61":1464,"_347":26},{"_369":1468,"_371":3293,"_373":1470,"_375":3294,"_377":1472},{"_369":3276,"_371":3277,"_373":3225,"_375":3278,"_377":3279},{"_369":3259,"_371":3260,"_373":3225,"_375":3261,"_377":3262},{"_369":3242,"_371":3243,"_373":3225,"_375":3244,"_377":3245},{"_369":3223,"_371":3224,"_373":3225,"_375":3226,"_377":3227},"e46da461",{"_416":3233,"_418":3234,"_862":3235,"_422":3236,"_371":584},"---\ndescription: \"Replies to every user prompt with original sea shanty lyrics and pirate-themed emojis, powered solely by Anthropic.\"\ntags: [\"sea-shanty\", \"pirate\", \"for-fun\"]\nauthors: []\ncapabilities: [\"sea-shanty-generation\"]\nintegrations:\n - \"anthropic\"\n---\n\nReplies to every user prompt with the lyrics of an original sea shanty, themed around the user's message and dressed up with pirate emojis. Uses only Anthropic (Claude) for generation.\n",{"_379":3228,"_381":3229,"_385":3230,"_387":1476,"_389":3231},"2026-05-01T17:42:29Z","Replies to every user prompt with original sea shanty lyrics and pirate-themed emojis, powered solely by Anthropic.",[1480,1481,1482],[1479],[3232],{"_59":568,"_61":569,"_397":24},{"_465":3241},{"_450":3238},{"_351":1489},{"_568":3237},{"_426":568},{"_454":3239},{"_456":3240},[462],"registry.astropods.ai/matt/pirate-parrot:e46da461","f1227841",{"_416":3250,"_418":3251,"_862":3252,"_422":3253,"_371":584},{"_379":3228,"_381":3246,"_385":3247,"_387":1476,"_389":3248},"2026-05-01T17:48:08Z",[1480,1481,1482],[1479],[3249],{"_59":568,"_61":569,"_397":24},{"_465":3258},{"_450":3255},{"_351":1489},{"_568":3254},{"_426":568},{"_454":3256},{"_456":3257},[462],"registry.astropods.ai/matt/pirate-parrot:f1227841","7d432bee",{"_416":3267,"_418":3268,"_862":3269,"_422":3270,"_371":584},{"_379":3228,"_381":3263,"_385":3264,"_387":1476,"_389":3265},"2026-05-16T20:14:57Z",[1480,1481,1482],[1479],[3266],{"_59":568,"_61":569,"_397":24},{"_465":3275},{"_450":3272},{"_351":1489},{"_568":3271},{"_426":568},{"_454":3273},{"_456":3274},[462],"registry.astropods.ai/matt/pirate-parrot:7d432bee","c01396dd",{"_416":3284,"_418":3285,"_862":3286,"_422":3287,"_371":584},{"_379":3228,"_381":3280,"_385":3281,"_387":1476,"_389":3282},"2026-07-10T23:52:18Z",[1480,1481,1482],[1479],[3283],{"_59":568,"_61":569,"_397":24},{"_465":3292},{"_450":3289},{"_351":1489},{"_568":3288},{"_426":568},{"_454":3290},{"_456":3291},[462],"registry.astropods.ai/matt/pirate-parrot:c01396dd",{"_416":3299,"_418":3300,"_862":3301,"_422":3302,"_371":584},{"_379":1473,"_381":3295,"_385":3296,"_387":1476,"_389":3297},[1480,1481,1482],[1479],[3298],{"_59":568,"_61":569,"_397":24},{"_465":1493},{"_450":3305},{"_351":1489},{"_642":3303},{"_422":3304,"_426":645},[647],{"_454":3306},{"_456":3307},[462],"serverAuth","serverTheme","dark","activeAccount"]