CASAN/.claude/agents/steps/steps-10-12-implement.md

Steps 10–12: Implementation & QA Phase

Boss MUST read this file before executing Steps 10–12. Protocols referenced: protocols/gate-retry-protocol.md, protocols/report-gate-protocol.md, protocols/implement-delegation.md

---

STEP 10 — Implementation + Build & Fix (Auto-Retry)

Key	Value
Agent	speckit.implement
Model	gpt-5-3-codex
Input	tasks.md, plan.md, data-model.md, contracts/, docs/output/ipa-docs/testcase/testcase-<mod-id>-<short-name>.md
Report	reports/10-implement-report.md (NN=10, phase=implement)
Gate	BUILD GATE (Auto-Retry) + REPORT HARD GATE (+ "Test Results" + "Screen Verification" sections required)
Max retries	5

Delegation: Per protocols/implement-delegation.md § STEP 10.

Additional context:

feature-id: <feature-id>
module-id: <mod-id>
report-nn: 10
report-phase: implement
pipeline-context: docs/output/output_logs/<feature-id>/pipeline-context.yaml

Phase 1 — Gen test:

• Every source file under src/ must have a dedicated test file under test/ or co-located (following standard TypeScript/Jest testing conventions)
• All checklist items in docs/output/specs/<feature-id>/checklists/ must be resolved before proceeding
• Auto-resolve any checklist ambiguities per protocols/auto-resolve-protocol.md

⛔ Test Quality Mandate (Post-Mortem P-08):

• Integration tests MUST use real database (SQLite for local, Docker DB for CI) — in-memory mocks are PROHIBITED
• Every endpoint test MUST verify: HTTP status, response shape, actual DB state change (read-back after write)
• Authorization tests MUST verify: unauthenticated → 401, wrong role → 403, ownership violation → 403
• Pseudo-tests (tests that always pass, mock everything, or have no assertions) are treated as CRITICAL review failure

Phase 2 — Implement:

• Execute all tasks in docs/output/specs/<feature-id>/tasks.md phase by phase
• All checklist items in docs/output/specs/<feature-id>/checklists/ must be resolved before proceeding
• Auto-resolve any checklist ambiguities per protocols/auto-resolve-protocol.md

Phase 3 — Build & Fix:

1. Fix all compile/lint errors (get_errors → fix → repeat until zero)
2. Build frontend: cd frontend && npm install && npm run build
3. Start Docker (if docker-compose.dev.yml exists): docker compose -f docker/docker-compose.dev.yml up -d
4. Build backend: cd backend && npm install && npm run build
5. Verify startup (if Docker available): cd backend && npm run start:dev

Gate logic:

• ✅ Build succeeds + app starts → proceed to Step 11
• ❌ Build/startup fails → Auto-Retry Loop:
  1. Capture error log
  2. Write [ISSUE] in boss log
  3. Invoke speckit.implement: "Fix build/startup errors: . Minimal fix."
  4. Retry build sequence
  5. If retry > 5: [ESCALATION], mark PARTIAL COMPLETE

⛔ [REPORT GATE] per protocols/report-gate-protocol.md

---

STEP 11 — Code Review (Auto-Retry)

Key	Value
Agent	okr.reviewcode
Model	claude-sonnet-4-6
Input	Implemented source code, spec, tasks, constitution
Report	reports/11-review-code-report.md
Gate	REVIEW GATE (Auto-Retry) + REPORT HARD GATE
Fix agent	speckit.implement
Max retries	5

Gate logic:

• ✅/⚠️ → proceed to Step 12
• ❌ REJECTED → invoke speckit.implement to fix CRITICAL issues → re-invoke okr.reviewcode
• Escalation after 5 retries → continue to Step 12

Additional DB Data Check: Code review MUST also verify:

• All screen data is fetched from the database (via Prisma Client / API endpoints), NOT from mock/hardcoded data
• Prisma seed script (backend/prisma/seed.ts) includes necessary seed data for the screens to display real content
• Frontend components call real API endpoints (not mock adapters or static JSON)
• If mock data is detected, mark as ❌ CRITICAL and instruct fix agent to replace with DB-backed data

⛔ [REPORT GATE] per protocols/report-gate-protocol.md

---

STEP 12 — Final QA Audit: Test Execution (Independent QA)

Key	Value
Agent	okr.testkit
Model	claude-sonnet-4-6
Mode	run-tests
Prerequisite	Step 10 PASSED
Report (Phase C)	reports/12-testkit-report.md
Report (Phase D)	docs/output/ipa-docs/testreport/testreport-<MOD-ID>-<short-name>.md
Gate	TEST GATE (BACK-TO-PLAN on FAIL)

⚠ This is the FINAL quality gate before launch. No screen shown until ALL tests pass.

Delegation $ARGUMENTS:

run-tests <feature-id>

Additional delegation context (mandatory):

Pipeline report (Phase C) MUST include:

1. ## Test Execution Summary — Category | Total | Passed | Failed | Skipped | Pass Rate
2. ## Failed Test Details — every failed test with TC-ID, failure reason, design reference
3. ## Retry Log — Retry Count | Target | Fix Applied | Result
4. ## Screen Verification Results (E2E) — per-screen accessibility
5. ## Coverage — Istanbul/c8 metrics
6. ## Overall Verdict — PASS / FAIL

IPA detail report (Phase D) generation rules:

• Take docs/output/ipa-docs/testcase/testcase-<MOD-ID>-<short-name>.md as BASE
• Rename title to Test Execution Result Report
• Keep ALL test case rows intact — fill Execution Result, Verdict, Notes columns only
• Add ## Test Execution Summary, ## Coverage Results at top
• Append ## Screen Verification Results, ## SRS/BD/DD Compliance Check, ## Overall Verdict at bottom

GATE — FAIL ← BACK TO PLAN (Full Fix Cycle):

Per protocols/gate-retry-protocol.md § BACK-TO-PLAN Fix Cycle:

• ✅ ALL tests PASS + coverage ≥ 80% → proceed to Step 13
• ❌ Tests FAIL → re-invoke pipeline from STEP 6 (max 3 full cycles)
• If 3 cycles exhausted → [ESCALATION], proceed to Step 13 with failure report

⛔ [REPORT GATE] — Both files required:

1. reports/12-testkit-report.md (Phase C)
2. docs/output/ipa-docs/testreport/testreport-<MOD-ID>-<short-name>.md (Phase D)