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

Fetch Meta to KB

Core Goal

• Pull journal-article records from Crossref after a given --from-date.
• Read ISSN seed rows from journals_issn (journal, issn1).
• Insert rows into journals with ON CONFLICT (doi) DO NOTHING.
• Keep the implementation aligned with fetch_meta_to_kb.py.

Run Workflow

1. Set database connection env vars (user-managed keys prefixed with KB_):

• KB_DB_HOST
• KB_DB_PORT
• KB_DB_NAME
• KB_DB_USER
• KB_DB_PASSWORD
• KB_LOG_DIR (required, log output directory)

2. Run incremental fetch with a required date:

python3 scripts/fetch_meta_to_kb.py --from-date 2024-05-01

• If executing through an exec tool call, set timeout to 1800 seconds (30 minutes).

3. Check logs in:

• ${KB_LOG_DIR}/fetch-meta-to-kb-YYYYMMDD-HHMMSS.log (UTC timestamp, one file per run)

4. Build user-facing summary strictly from the current run output:

• Prefer RUN_SUMMARY_JSON emitted by fetch_meta_to_kb.py.
• If JSON is unavailable, parse only this run's ${KB_LOG_DIR}/fetch-meta-to-kb-YYYYMMDD-HHMMSS.log.
• total_inserted must mean rows inserted in this run (after DOI dedup), not cumulative rows in table.

Behavior Contract

• Query Crossref endpoint: https://api.crossref.org/journals/{issn}/works.
• Filter with type:journal-article,from-pub-date:<from-date>.
• Keep only items whose container-title equals target journal title (case-insensitive).
• Continue pagination with cursor until no matching items remain.
• Store fields in journals: title, doi, journal, authors, date, abstract (nullable when Crossref has no abstract).
• Reporting/announcement metrics must use current-run log/summary only.
• Do not compute announcement counts via database-wide or time-window SQL such as WHERE date >= ....

Scope Boundary

• Implement only Crossref incremental fetch + insert into journals.

Script

• scripts/fetch_meta_to_kb.py