Imported from tiangong-ai/agent-skills (ai-tech-fulltext-fetch/SKILL.md). Install upstream with npx skills add tiangong-ai/agent-skills --skill ai-tech-fulltext-fetch. Copyright stays with the author.

AI Tech Fulltext Fetch

Core Goal

• Reuse the same SQLite database populated by ai-tech-rss-fetch.
• Fetch article body text from each RSS entry URL.
• Persist extraction status and text in a companion table (entry_content).
• Support incremental runs and safe retries without creating duplicate fulltext rows.

Triggering Conditions

• Receive a request to fetch article body/full text for entries already in ai_rss.db.
• Receive a request to build a second-stage pipeline after RSS metadata sync.
• Need a stable, resumable queue over existing entries rows.
• Need URL-based fulltext persistence before chunking, indexing, or summarization.

Workflow

1. Ensure metadata table exists first.

• Run ai-tech-rss-fetch and populate entries in SQLite before using this skill.
• This skill requires the entries table to exist.
• In multi-agent runtimes, pin DB to the same absolute path used by ai-tech-rss-fetch:

export AI_RSS_DB_PATH="/absolute/path/to/workspace-rss-bot/ai_rss.db"

2. Initialize fulltext table.

python3 scripts/fulltext_fetch.py init-db --db "$AI_RSS_DB_PATH"

3. Run incremental fulltext sync.

• Default behavior fetches rows that are missing full text or currently failed.

python3 scripts/fulltext_fetch.py sync \
  --db "$AI_RSS_DB_PATH" \
  --limit 50 \
  --timeout 20 \
  --min-chars 300

4. Fetch one entry on demand.

python3 scripts/fulltext_fetch.py fetch-entry \
  --db "$AI_RSS_DB_PATH" \
  --entry-id 1234

5. Inspect extracted content state.

python3 scripts/fulltext_fetch.py list-content \
  --db "$AI_RSS_DB_PATH" \
  --status ready \
  --limit 100

Data Contract

• Reads from existing entries table:
  • id, canonical_url, url, title.
• Writes to entry_content table:
  • entry_id (unique, one row per entry)
  • source_url, final_url, http_status
  • extractor (trafilatura, html-parser, or none)
  • content_text, content_hash, content_length
  • status (ready or failed)
  • retry_count, last_error, timestamps.

Extraction and Update Rules

• URL source priority: canonical_url first, fallback to url.
• Attempt trafilatura extraction when dependency is available, fallback to built-in HTML parser.
• Upsert by entry_id:
  • Success: write/update full text and reset retry_count to 0.
  • Failure with existing ready content: keep old text, keep status ready, record last_error.
  • Failure without ready content: status becomes failed, increment retry_count, set next_retry_at.
• Failed retries are capped by --max-retries (default 3) and paced by --retry-backoff-minutes.
• --force allows refetching already ready rows.
• --refetch-days N allows refreshing rows older than N days.

Configurable Parameters

• --db
• AI_RSS_DB_PATH (recommended absolute path in multi-agent runtime)
• --limit
• --force
• --only-failed
• --refetch-days
• --oldest-first
• --timeout
• --max-bytes
• --min-chars
• --max-retries
• --retry-backoff-minutes
• --user-agent
• --disable-trafilatura
• --fail-on-errors

Error Handling

• Missing entries table: return actionable error and stop.
• Network/HTTP/parse errors: store failure state and continue processing other entries.
• Non-text content types (PDF/image/audio/video/zip): mark as failed for that entry.
• Extraction too short (--min-chars): treat as failure to avoid low-quality body text.

References

• references/schema.md
• references/fetch-rules.md

Assets

• assets/config.example.json

Scripts

• scripts/fulltext_fetch.py