Daily Content Curator

Curates personalized content based on your interests. Supports two modes:

• Standalone — works locally, no external account needed
• Eir — full AI-powered curation with heyeir.com delivery

---

Standalone Mode

A complete local curation pipeline. No Eir account required.

Flow

1. Extract interests    → scan conversations, save to config/interests.json
2. Search              → Grounding API + RSS for each interest topic
3. Select + Crawl      → LLM picks best candidates, fetches full content
4. Generate            → LLM writes structured summaries to data/output/
5. Digest              → Compile daily digest from generated items

Quick Start

Step 1: Configure search provider

Edit config/settings.json:

{
  "mode": "standalone",
  "search": {
    "grounding_base_url": "https://api.your-provider.com/v3",
    "grounding_api_key": "YOUR_KEY"
  }
}

Recommended providers: Brave Search API, Tavily API, or any compatible grounding service.

Step 2: Set up interests

Option A — Manual: create config/interests.json:

{
  "topics": [
    {"label": "AI Agents", "keywords": ["autonomous agents", "tool use", "agent frameworks"]},
    {"label": "Prompt Engineering", "keywords": ["prompting", "chain-of-thought"]}
  ],
  "language": "en",
  "max_items_per_day": 8
}

Option B — Auto-extract from conversations (agent-driven):

Read references/interest-extraction-prompt.md, then scan recent conversations
and write discovered topics to config/interests.json

Step 3: Run the pipeline

# Full pipeline (agent-driven, recommended)
# Agent reads this SKILL.md, runs each step in sequence

# Or run individual steps:
python3 -m pipeline.search              # search for each topic
python3 -m pipeline.candidate_selector  # prepare topics for judgment
python3 -m pipeline.crawl               # fetch full content
python3 -m pipeline.pack_tasks          # bundle into task files

# Generate is agent-driven (LLM writes content from task files)
# Digest is agent-driven (LLM compiles daily summary)

Step 4: Schedule daily cron

openclaw cron add --name "daily-curate" \
  --cron "0 8 * * *" --tz "Asia/Shanghai" \
  --session isolated \
  --message "Run the standalone content curation pipeline from eir-daily-content-curator skill. Follow SKILL.md Standalone Mode steps: search → select → crawl → pack → generate → digest. Deliver the digest summary to me."

Output

Generated content is saved to data/output/{date}/:

data/output/2026-04-20/
  meta-layoffs-ai.json          # individual content items
  china-ai-regulation.json
  digest.md                     # compiled daily digest

Each item follows this format:

{
  "title": "...",
  "summary": "2-3 sentences",
  "body": "2-4 paragraphs with analysis",
  "sources": [{"name": "...", "url": "...", "published": "..."}],
  "topic": "AI Industry",
  "generated_at": "2026-04-20T08:30:00Z"
}

Digest Format

The daily digest is a markdown file combining all items:

# Daily Digest — 2026-04-20

## 🔴 Focus
- **Meta cuts 8,000 jobs for AI pivot** — ...

## 🟡 Interesting
- **China bans AI companions for minors** — ...

## 🌱 Seeds
- **New prompt engineering benchmark** — ...

---

Eir Mode

Full AI-powered curation with delivery to the Eir app (heyeir.com).

Flow

1. Fetch directives    → GET /oc/curation (topics + search hints from API)
2. Search              → Grounding API + SearXNG for each directive
3. Select + Crawl      → LLM picks candidates, fetches full content
4. Pack tasks          → Bundle into self-contained task files
5. Generate + POST     → LLM writes content, POST to Eir Content API

Setup

1. Connect Eir account: node scripts/connect.mjs <PAIRING_CODE>
2. Configure config/eir.json with API credentials
3. Set "mode": "eir" in config/settings.json

Architecture (2-Job Split)

Job A: eir-material-prep (07:00)
  search → select → crawl → pack
  Output: data/v9/tasks/{content_slug}.json

Job B: eir-content-gen (07:45)
  For each task → spawn subagent → generate → POST
  Output: content posted to Eir API

Cron Setup

# Material preparation
openclaw cron add --name "eir-material-prep" \
  --cron "0 7 * * *" --tz "Asia/Shanghai" \
  --session isolated \
  --message "Run eir-daily-content-curator material prep: search → select → crawl → pack tasks."

# Content generation
openclaw cron add --name "eir-content-gen" \
  --cron "45 7 * * *" --tz "Asia/Shanghai" \
  --session isolated \
  --message "Run eir-daily-content-curator content generation: read task files, generate content, POST to API."

Content Quality Rules

• dot.hook ≤10 CJK chars / ≤6 EN words
• dot.category: focus | attention | seed | whisper
• l1.bullets 3-4 items, each ≤20 CJK chars
• sources must have at least 1 entry
• Never set any field to null — use "" or []
• Only generate zh (no translation)

See references/content-spec.md for full field constraints. See references/writer-prompt-eir.md for the generation prompt.

Validation

python3 -m pipeline.validate_content           # check all generated files
python3 -m pipeline.validate_content --fix     # auto-fix common issues

---

Shared Components

Both modes use these pipeline modules:

Module	Purpose
pipeline/search.py	Search via Grounding API (Brave/Tavily), SearXNG fallback
pipeline/crawl.py	Fetch full content via Grounding Browse, Crawl4AI fallback
pipeline/grounding.py	Generic grounding API client (configurable baseURL + key)
pipeline/candidate_selector.py	Group results by topic, prepare for LLM selection
pipeline/pack_tasks.py	Bundle candidates into self-contained task files
pipeline/config.py	Shared configuration and path resolution
pipeline/date_extractor.py	Extract publish dates from HTML

Search Fallback Chain

Grounding API (primary) → SearXNG (fallback) → Crawl4AI/web_fetch (content)

Configuration

config/settings.json:

{
  "mode": "standalone",
  "search": {
    "grounding_base_url": "https://api.your-provider.com/v3",
    "grounding_api_key": "YOUR_KEY",
    "searxng_url": "http://localhost:8888",
    "crawl4ai_url": "http://localhost:11235"
  },
  "max_items_per_day": 8
}

• grounding_base_url + grounding_api_key: Primary search provider
• searxng_url: Optional local SearXNG instance (fallback)
• crawl4ai_url: Optional local Crawl4AI instance (fallback)

---

Interest Management

Standalone: Local Interest Extraction

The agent extracts interests from recent conversations and saves to config/interests.json. See references/interest-extraction-prompt.md for extraction rules.

{
  "topics": [
    {"label": "AI Agents", "keywords": ["autonomous agents", "tool use"], "freshness": "7d"},
    {"label": "华为汽车", "keywords": ["鸿蒙智行", "问界"], "freshness": "3d"}
  ],
  "language": "zh",
  "max_items_per_day": 8
}

Eir: API-Synced Interests

Interests are managed via the Eir API (GET /oc/curation returns directives with search hints). See references/eir-interest-rules.md for tier guidelines.

---

RSS Sources

Both modes can supplement search with RSS feeds. Configure in config/sources.json:

{
  "rss": [
    {"name": "Techmeme", "url": "https://www.techmeme.com/feed.xml", "rating": "S", "lang": "en"},
    {"name": "TechCrunch", "url": "https://techcrunch.com/feed/", "rating": "A", "lang": "en"}
  ]
}

Ratings: S (check every 4h), A (8h), B (24h).

---

References

File	Contents
references/content-spec.md	Field types, limits, validation rules
references/eir-api.md	Eir API endpoints and payloads
references/writer-prompt-eir.md	Eir mode content generation prompt
references/writer-prompt-standalone.md	Standalone mode generation prompt
references/eir-interest-rules.md	Curation tier guidelines
references/interest-extraction-prompt.md	Interest extraction from conversations

---

Quick Reference

Task	Command
Search	python3 -m pipeline.search
Search (single topic)	python3 -m pipeline.search --topic ai-health
Select candidates	python3 -m pipeline.candidate_selector
Crawl	python3 -m pipeline.crawl
Pack tasks	python3 -m pipeline.pack_tasks
Validate	python3 -m pipeline.validate_content
Standalone curate (legacy)	python3 scripts/standalone/curate.py