favorites-curator（最爱策展人）

Favorites Curator

Maintain a file-backed catalog of useful local resources.

Do First

1. Run scripts/scan_favorites.py to refresh entries and the latest snapshot.
2. Run scripts/daily_digest.py to compare the latest snapshot with the previous one and produce a digest.
3. Run scripts/notify_candidates.py when the user wants short notification copy for notable items.

Paths

• Entries: favorites/entries/
• Snapshots: favorites/snapshots/
• Reports: favorites/reports/
• Cache: favorites/enrichment-cache.json
• Skill code: skills/favorites-curator/
• Naming and schema notes: references/naming-and-model.md

Covered Sources

• ~/ai git repositories
• brew installed formulae and casks
• /Applications and ~/Applications
• ~/.codex/skills
• ~/.claude/skills
• ~/.agents/skills
• ~/.openclaw/workspace/skills
• ~/.openclaw/extensions
• ~/.openclaw/hooks

Expected Workflow

Refresh catalog

python3 skills/favorites-curator/scripts/scan_favorites.py

Use --limit-source to iterate on one source:

python3 skills/favorites-curator/scripts/scan_favorites.py --limit-source brew

Generate daily digest

python3 skills/favorites-curator/scripts/daily_digest.py

This produces a Markdown report with:

• new entries
• updated entries
• likely similar items worth deduping or cross-referencing
• notable items worth surfacing

Emit notification copy

python3 skills/favorites-curator/scripts/notify_candidates.py

Use --top N to cap the number of lines.

Data Rules

• Prefer local metadata over online enrichment.
• Leave missing fields empty or mark them as inferred in field_sources.
• Keep filenames ASCII, sortable, and machine-friendly.
• Store one entry per file.
• Update the same entry file on re-scan instead of creating duplicates.

Output Contract

Each entry file must keep these fields in frontmatter:

• name
• author
• data_type
• install_path
• install_time
• update_time
• summary
• simple_usage
• source_url

Other helper fields are allowed when useful for automation.

Notes

• The scripts are local-first and safe to run repeatedly.
• brew info --json=v2 --installed is used once per scan; avoid tight cron loops.
• Read references/naming-and-model.md before changing filename rules or schema.