---

name: purefeed description: "Monitors Twitter/X feeds with AI signal detection. Searches tweets semantically, manages signal detectors, generates human-sounding posts, checks AI detection scores, and publishes/schedules content to X via Typefully. Use when the user wants to browse their Twitter feed, find tweets about a topic, set up content monitoring, draft tweets from curated content, or publish to X/Twitter. Do NOT use for general Twitter browsing without a Purefeed account." user-invocable: true allowed-tools: ["bash"] metadata: {"openclaw":{"requires":{"bins":["curl"],"env":["PUREFEED_API_KEY"]},"primaryEnv":"PUREFEED_API_KEY","emoji":"📡"}}

Purefeed

API Base: https://www.purefeed.ai/api/v1 Auth: Authorization: Bearer $PUREFEED_API_KEY

Output Rules

Follow these rules for EVERY response that includes tweet data:

1. Format ALL screen names as clickable links: [@screen_name](https://x.com/screen_name). Never output plain @screen_name.
2. Format ALL tweet references as clickable links: [Tweet](https://x.com/screen_name/status/tweet_id).
3. Include view count with eye emoji: 👁 81K.
4. Sort by views or engagement descending unless user requests otherwise.

Example output line:

[@CryptoAyor](https://x.com/CryptoAyor) 👁 81K — detailed thread about $JELLY manipulation

Setup

1. Get API key at purefeed.ai/profile → Public API Keys → Create Key
2. Set it: openclaw config set skills.purefeed.env.PUREFEED_API_KEY "pf_live_YOUR_KEY"
3. Verify: curl -s https://www.purefeed.ai/api/v1/auth/me -H "Authorization: Bearer $PUREFEED_API_KEY"

Tool Dependencies

list styles ──→ styleId ──→ generate post
list signals ──→ signal_id ──→ get/update/delete signal, get signal matches
list folders ──→ folder_id ──→ get/create/rename/delete folder, add/remove tweets
get feed / search / signal matches ──→ tweet_ids ──→ generate post / get signal insights
generate post ──→ draft text ──→ humanize post ──→ publish / schedule / draft / queue

Always list styles before generating a post. Always list signals before signal-specific calls.

How to Find Tweets by Topic

Follow this exact sequence when the user asks "what's new about X?" or "find tweets about Y".

Step 1 — Find a matching active signal

GET /signals?search=TOPIC&active=true

The search parameter uses semantic/vector search — search=ai finds "Artificial intelligence", "AI Research", etc. If empty, try broader terms or GET /signals?active=true to see all active signals.

Step 2 — Get matches from that signal

GET /signals/{id}/matches?limit=20

Signal matches are the primary data source. They include AI-generated analysis (sentiment, category, insights). Do NOT skip to feed search.

Step 3 — Fall back to feed search ONLY if no signal found

GET /feed?limit=20&search=TOPIC
POST /search → {"query": "topic description", "limit": 20}

Step 4 — Filter feed by signal IDs (optional)

GET /feed?signal_ids=236,237&limit=20

Workflows

Find and share tweets

1. Find tweets using the strategy above
2. GET /styles — pick a writing style
3. POST /studio/generate — generate post from tweet IDs + styleId
4. POST /studio/humanize — if score ≤ 30 use as-is, if > 30 use rewrite
5. POST /studio/publish — publish to X

Set up monitoring

1. POST /signals — create signal with name + description + tags + color + cron + timezone (auto-activates)
2. Wait up to 6 hours for processing
3. GET /signals/{id}/matches — check results
4. PUT /signals/{id} — refine description if too many irrelevant matches

First run

1. GET /auth/me — verify API key
2. GET /styles — cache writing styles
3. GET /signals — see signal configurations
4. GET /folders — see bookmark folders

Key Concepts

• Signal: AI content detector with a name + description. Active if signals_subscriptions is non-empty; inactive if []. When creating: always set tags and color, never set include_keywords unless user explicitly asks (they are very restrictive).
• Writing Style: Named instruction set for post generation. Must call GET /styles to get valid styleId before generating.
• Typefully: Publishing backend. Required for publish/schedule/draft/queue. If not connected, tell user to set it up in Purefeed settings.

Error Handling

Error	Agent Action
401 Unauthorized	Tell user to create new key at purefeed.ai/profile
429 Too Many Requests	Wait and retry. Check Retry-After header
"No Typefully connection"	Tell user to connect Typefully in Purefeed settings
"No LLM API keys configured"	Tell user to add LLM API keys in Studio settings
"Writing style not found"	Call GET /styles to get valid IDs
"Signal not found"	Call GET /signals to get valid IDs

API Reference

Read API_REFERENCE.md for full endpoint documentation, parameters, curl examples, and response shapes.

All endpoints return { "data": ..., "error": null } on success and { "data": null, "error": { "message": "...", "code": "..." } } on error. Streaming endpoints return text/plain.

Endpoint Summary

Method	Path	Purpose
GET	/auth/me	Verify API key
GET	/feed	Tweets ranked by signal relevance
POST	/search	Full-text search across matched tweets
GET	/feed/signals	AI signal analysis for specific tweet IDs
GET	/folders	List bookmark folders
POST	/folders	Create a folder ({ "name": "..." })
PATCH	/folders/:id	Rename a folder ({ "name": "..." })
DELETE	/folders/:id	Delete a folder and its items
GET	/folders/:id/tweets	Tweets in a folder
POST	/folders/:id/tweets	Add tweet to folder ({ "tweet_id": "..." })
DELETE	/folders/:id/tweets?tweet_id=X	Remove tweet from folder
GET	/signals	List signals (supports semantic search)
POST	/signals	Create + auto-activate a signal
GET	/signals/:id	Signal details
PUT	/signals/:id	Update signal
DELETE	/signals/:id	Delete signal (irreversible)
GET	/signals/:id/matches	Tweets matching a signal
GET	/styles	List writing styles
POST	/studio/generate	Generate tweet/thread (streaming)
POST	/studio/humanize	AI detection score + rewrite (streaming)
POST	/studio/publish	Publish to X immediately
POST	/studio/schedule	Schedule for specific time
POST	/studio/draft	Save as Typefully draft
POST	/studio/queue	Add to next posting slot

Rate Limits

• 60 requests/minute per API key
• Streaming endpoints count as 1 request each
• 429 responses include Retry-After header