flight-ghost-harvest-skill

Ctrip flight harvesting skill for logged-in desktop Chrome sessions, with browser-driven collection, SQLite caching, and conservative planner defaults for domestic and international routes.

What This Repo Does

• drives the real Ctrip browser flow instead of Playwright or direct HTTP scraping
• harvests one-way and round-trip flight result pages from desktop Chrome
• supports route freshness checks through a shared SQLite cache
• exports raw and normalized JSON for downstream analysis
• includes Claude-compatible agent metadata under .claude/agents/

Key Files

• SKILL.md: primary skill instructions
• .claude/agents/flight-ghost-harvest.md: Anthropic-compatible agent entrypoint
• scripts/plan_harvest_run.py: unified planner for freshness-aware harvest runs
• scripts/harvest_ctrip_chrome.py: browser-side Ctrip result extraction
• scripts/check_db_freshness.py: route freshness lookup
• scripts/sync_flights_db.py: SQLite sync for normalized output
• references/database-guide.md: how the agent should read the shared database

Typical Flow

1. Ask the user for route and, unless they explicitly want one-way only, approximate stay length.

2. Check the shared SQLite cache:

   python3 scripts/check_db_freshness.py SHA TYOA

3. Plan or execute the harvest:

   python3 scripts/plan_harvest_run.py SHA TYOA --stay-days 5
   python3 scripts/plan_harvest_run.py SHA TYOA --stay-days 5 --execute

4. Inspect raw or normalized output under out/.

Notes

• The planner defaults to conservative browser actions: --expand-limit 0 and --scroll-steps 0.
• The shared SQLite database lives at ~/.local/share/flight-harvest/flights.db.
• For direct database inspection, start with references/database-guide.md.