QA Plan

Generates structured QA plan for sprint/feature/story. Reads story files + GDDs, classifies by test type, produces plan with automated tests, manual checks, smoke scope, playtest requirements. Run before sprint begins.

Output: production/qa/qa-plan-[sprint-slug]-[date].md

Phase 1: Parse Scope

• sprint — read latest file in production/sprints/, extract story paths (prefer sprint-status.yaml if exists)
• feature: [name] — glob production/epics/*/story-*.md, filter by system name
• story: [path] — single file
• No argument — AskUserQuestion: Current sprint / Specific feature / Specific story / Full epic

Missing story files → note as MISSING, continue with rest.

Phase 2: Load Inputs

Per story extract: title/ID, story type, acceptance criteria, implementation files, engine notes, GDD/ADR references, estimate, dependencies.

Supporting context (once): design/gdd/systems-index.md, GDD Acceptance Criteria + Formulas sections for each referenced GDD, docs/architecture/control-manifest.md forbidden patterns.

Phase 3: Classify Stories

Story Type	Indicators
Logic	Calculations, formulas, thresholds, state transitions, data validation
Integration	≥2 systems interacting, events across boundaries, save/load, network
Visual/Feel	Animation, VFX, shaders, timing, screen shake, audio sync
UI	Menus, HUD, buttons, screens, tooltips, panels
Config/Data	Balance tuning, data files, configuration only

Mixed stories: assign primary by highest implementation risk, note secondary. Show classification summary table before Phase 4.

Phase 4: Generate Test Plan

Sections:

• Test Summary — table: story, type, automated test required, manual verification required
• Automated Tests Required — per story: test file path, what to test (formulas/rules/transitions), edge cases (zero/max/invalid + GDD edge cases), estimated test count. No GDD formula → flag gap
• Manual QA Checklist — per story: verification method, sign-off (designer/lead/qa), evidence to capture, concrete checklist items. Subjective criteria → supplemented with benchmark
• Smoke Test Scope — launch, new session, primary mechanic, regression risks, save/load, performance
• Playtest Requirements — table: story, goal (question to answer), min sessions, target player type. Notes path for sign-off
• Definition of Done — all ACs verified, test files exist (Logic/Integration), manual evidence (Visual/UI), smoke passes, no regressions, code reviewed, story status updated

Use actual story titles and GDD content — no placeholder text.

Phase 5: Write Output

Show plan (or summary if very long). "May I write to production/qa/qa-plan-[sprint-slug]-[date].md?"

After writing: share with team before sprint, run /smoke-check sprint after implementation, create test files at listed paths before marking done.

Protocol

• Never write without asking
• Classify conservatively: ambiguous Logic/Integration → Integration (needs both test types)
• Don't invent test cases beyond ACs and GDD formulas — flag gaps instead
• Playtest requirements are advisory — user decides borderline cases
• AskUserQuestion only for scope selection; rest is present-then-approve

Checklist

• Scope parsed correctly (sprint/feature/story files found and loaded)
• Stories classified by test type with primary/secondary assignments
• Coverage mapped: every AC has at least one automated or manual verification
• Test plan written to production/qa/qa-plan-[sprint-slug]-[date].md
• Edge cases identified from GDD formulas and flagged where missing
• Playtest requirements include target player type and minimum sessions