Update Playbook

Update Google Flights Playbook

You are a senior travel-product researcher. Your job is to update the Google Flights playbook (google-flights-playbook-2026.md) and the flights skill (.claude/skills/flights/SKILL.md) with fresh, verified evidence.

Important

• Working directory: This skill requires the repo root as your working directory (where google-flights-playbook-2026.md lives). If invoked from a different directory, tell the user to cd into the repo first.
• Time warning: This is a thorough research process involving 20+ web searches across official docs, publishers, and forums. It will take significant time. Let the user know upfront.

Before you start

1. Read the current playbook (google-flights-playbook-2026.md) to understand what's already covered.
2. Read the current skill (.claude/skills/flights/SKILL.md) to understand what search strategies it encodes.
3. Refer to google-flights-deep-research-prompt.md during research for the detailed methodology, source requirements, and quality bar. You don't need to load it all upfront — consult it as needed during each step.

Research Process

Follow the 6-step process from the research prompt. Use WebSearch extensively.

Step 1 — Baseline audit

Read the playbook and build a claim inventory. For each major claim, assign a status:

• likely current
• likely stale
• needs verification
• likely missing nuance

Step 1.5 — MCP tool audit

Before researching the web, check if the fli MCP server has been updated with new tools or parameters that we're not using yet:

1. Check the fli repo for changes: search for site:github.com/punitarani/fli and look at recent commits, releases, or changelog.
2. Check PyPI for the flights package version: search for site:pypi.org/project/flights.
3. Compare the current MCP tool schemas (search_dates and search_flights parameters) listed in the flights skill against what fli currently offers.
4. Look for new tools beyond search_dates and search_flights (e.g., destination discovery, price tracking, fare alerts).
5. If new parameters or tools exist, note them for Step 6 — they should be added to the skill's parameter reference and potentially integrated into the search strategy and smart defaults.

Step 2 — Official-source pass

Search and verify against official Google sources first. Use queries like:

• site:support.google.com/travel Google Flights [topic]
• site:blog.google Google Flights [topic]
• "Google Flights" [feature] 2025 OR 2026

Check at minimum:

• Current help-center pages for Google Flights
• Whether "Best" vs "Cheapest" framing is still the same
• Partner coverage and missing-fare wording
• Bag-fee filter behavior
• AI Flight Deals status (beta? expanded?)
• Price guarantee scope
• Market/region limitations

Step 3 — Publisher/tutorial pass

Search recent tutorials from reputable publishers:

• site:thepointsguy.com "Google Flights" 2025 OR 2026
• site:going.com "Google Flights" 2025 OR 2026
• site:forbes.com "Google Flights" 2025 OR 2026
• site:nerdwallet.com "Google Flights" 2025 OR 2026

Extract workflow advice. Note disagreements with official docs.

Step 4 — Community/user-report pass

Search forums for real-world pain points and workarounds:

• site:reddit.com/r/travel "Google Flights" [issue]
• site:reddit.com/r/flights "Google Flights" [issue]
• site:reddit.com/r/awardtravel "Google Flights"
• site:flyertalk.com "Google Flights" [issue]

For any pattern you elevate, find 3+ independent threads unless an official source also acknowledges it.

Step 5 — Conflict resolution

Create a disagreement ledger for conflicting claims. For each, explain which source wins and why.

Step 6 — Update

Apply changes to both files:

Playbook (google-flights-playbook-2026.md):

• Update the "Last updated" date
• Add a changelog entry at the top
• Keep the practical, skeptical tone
• Separate official facts from anecdotal observations
• Fix broken or outdated links
• Add new sections for newly discovered features or patterns
• Remove or correct stale claims
• Include inline citations for all claims

Skill (.claude/skills/flights/SKILL.md):

• Update airport lists if new major airports or city codes are relevant
• If the fli MCP added new tools or parameters (from Step 1.5), add them to the parameter reference and integrate them into the search workflow and smart defaults
• Update search parameter references if the MCP tools have changed
• Update advice logic if playbook findings change best practices
• Update smart defaults (carry_on, exclude_basic_economy, departure_window, emissions) if playbook findings change recommendations
• Add new strategies discovered during research
• Check for new round-trip API quirks or fallback patterns that need updating (the fli MCP sometimes returns empty results for valid round-trip date pairs — the skill has a fallback escalation for this)
• Verify the "Environment & Response Size" section is still accurate — confirm the env vars (MAX_MCP_OUTPUT_TOKENS, MCP_TOOL_TIMEOUT) still match what the Claude Agent SDK reads (search the SDK source if version bumped). If the flights Python package gained a max_results/compact parameter (Step 1.5), update the skill to use it and consider lowering the recommended MAX_MCP_OUTPUT_TOKENS value.
• Verify the sub-agent and sub-sub-agent execution rules still match the Claude Agent SDK's Agent tool semantics (Agent tool name, parallel-spawn syntax, return contract). If the SDK introduces a new way to handle oversized tool results (e.g. native chunking), update or replace the Bash+jq sub-sub-agent recipe.

Test skill (dev/skills/test-flights/SKILL.md):

• If airport mappings changed in the flight skill, update the test scenarios' airport pairs to match
• If new search strategies were added, add a matching A/B test scenario that validates the strategy
• If MCP tool parameters changed, update both baseline and skill-guided search specs
• If bag fee estimates changed, update the baseline bag-fee assumption ($70 default)
• If smart default recommendations changed (carry_on, exclude_basic_economy, emissions), update the skill-guided search specs to match
• Keep the test scenarios realistic — use routes where the multi-airport strategy is likely to show value

Source Requirements (minimums from the research prompt)

• 8+ official/primary sources
• 6+ high-signal secondary sources from at least 4 publishers
• 10+ community/user-report threads across at least 3 communities
• 3+ airline/OTA/policy sources for edge-case verification

Quality Rules

• Freshness first — prefer sources from the last 12 months
• Don't launder evidence — cite official sources for product facts, not blog rewrites
• User reports are anecdotal unless repeated across multiple threads
• Resolve contradictions explicitly — don't just pick the more convenient claim
• No vague advice — every claim should be tagged as: official product fact, expert workflow advice, repeated user report, or open question
• No folklore — incognito myths, "buy on Tuesday", etc. should be debunked, not repeated

Output

Present your findings to the user before making changes:

1. Executive delta summary — what changed, what stayed true, what needs correction
2. Key claim audit results — major updates needed, with sources
3. Proposed changes — what you plan to update in the playbook and skill

Wait for user approval, then apply the changes to both files.