feat: Skywatch.brief — AIM 7-1-5 weather brief composition layer

[jayrav13]

Summary

• Adds Skywatch.brief(airport:) and skywatch brief AIRPORT — a single composition layer over existing Briefer + Nimbus primitives that produces an AIM 7-1-5-shaped weather brief for a single airport.
• Six slots fillable in MVP (current_conditions, destination_forecast, winds_aloft, adverse_conditions, vfr_not_recommended, afd), four hardcoded-unavailable per spec (synopsis, enroute_forecast, notams, atc_delays), one supplementary (afd).
• METAR is the only hard-fail — every other source degrades to { available: false, reason: ... } via the per-slot wrap helper. Adverse-conditions slot tracks per-source partial_failures independently.
• Two new infra additions: AirportLocator (METAR-derived coordinates + WFO lookup, cached 1 day) and AdverseFilter (polygon-intersect + distance + urgency partition).
• Bundled validation report against the canonical 7-question protocol (3 scenarios, naive-context LLM run): PASS — Q1–Q6 substantively correct on all 3 scenarios, Q7 anti-hallucination clean on all 3. See docs/superpowers/specs/2026-04-27-skywatch-brief-validation-results.md.

Spec: docs/superpowers/specs/2026-04-27-skywatch-brief-design.md.
Plan: docs/superpowers/plans/2026-04-27-skywatch-brief.md.

Validation thesis confirmed

The thesis — can an LLM produce a useful 1-800-WX-BRIEF-quality answer from current Skywatch primitives? — is confirmed for the single-airport MVP scope. The AIM-7-1-5 envelope shape, uniform { available, ... } slot contract, and per-slot reason strings did the load-bearing work: when slots were unavailable, the LLM degraded honestly using slot-adjacent data (AFD as synopsis substitute, PIREP temperatures as winds-aloft proxies) without inventing missing facts. KMSY exercised the system hardest — a SPECI METAR captured a 30-minute drop to IFR with -RA, and the adverse-conditions slot's polygon-intersect logic correctly surfaced both convective SIGMET 74C and an urgent low-level wind-shear PIREP at the field.

Bug surfaced during validation

AIRMET ingest fails uniformly with RGeo::Error::InvalidGeometry: Self-intersection, captured cleanly under partial_failures on every brief — exactly the degrade-gracefully behavior the spec required. Filed as #11.

Follow-up issues (deferred from MVP)

• #5 — Route input
• #6 — Coordinate input
• #7 — ETD-aware TAF/winds windowing
• #8 — AFD synopsis extraction
• #9 — Parallelize source fetches
• #10 — Human-readable text formatter
• #11 — AIRMET self-intersection (filed during validation)

Test plan

• bundle exec rspec — 414 examples, 0 failures
• bundle exec rubocop — 108 files, 0 offenses
• exe/skywatch brief KCDW — JSON to stdout
• exe/skywatch brief KACK — IFR scenario, JSON to stdout
• exe/skywatch brief KMSY — convective + IFR scenario, JSON to stdout
• Validation report committed with PASS verdict and judgment tables for all 3 scenarios
• All 7 follow-up issues filed (Skywatch.brief: support route input (from:/to:) #5–AIRMET ingest fails with RGeo::Error::InvalidGeometry: Self-intersection #11) and linked from the validation report

🤖 Generated with Claude Code