Flight Fare Watch

3. Collect public fares:

• Use browser-based collection (for example with local playwright skill) or user-provided exports.
• Store findings in public_fares.csv with source URL and timestamp.
• Follow references/public-source-playbook.md.

4. Compare with internal quotes:

• Run scripts/compare_quotes.py.
• Produce recommendation CSV + markdown summary.

5. Generate recommendations:

• Use policy from references/recommendation-rules.md.
• Always report recommendation type and measurable reason (price delta, schedule shift, stop count).

6. Publish daily report:

• Include route-level status: save-money, same-price-better, time-shift, monitor, or no-data.
• Include concrete alternative offer link and expected saving.

Recommendation Categories

Use exactly these categories:

• save-money: Public fare is materially cheaper than internal quote.
• same-price-better: Price is similar but quality is better (airline/stops/time window).
• time-shift: Small departure or return time shift gives meaningful savings.
• monitor: No strong alternative now; keep tracking.
• no-data: Missing comparable public data.

Command Patterns

If Python is not globally available, run scripts with uv:

uv run --python 3.11 scripts/build_watchlist.py \
  --internal data/internal_quotes.csv \
  --out data/watchlist.csv

uv run --python 3.11 scripts/compare_quotes.py \
  --internal data/internal_quotes.csv \
  --public data/public_fares.csv \
  --out-report outputs/daily-fare-report.md \
  --out-csv outputs/recommendations.csv

Report Contract

Return:

1. Output file paths.
2. Count by recommendation category.
3. Top opportunities by saving amount.
4. Data quality warnings (missing fields, unmatched routes, stale collection time).
5. Explicit non-goals (no purchase action, no approval write-back).

Guardrails

• Respect target site terms and robots guidance.
• Avoid aggressive crawling or bypass behavior.
• Keep recommendation language advisory, not guaranteed outcomes.
• Prefer reproducible runs from saved CSV snapshots.