Execute Implementation

/execute-implementation — GREEN Loop Phase

$TM = python3 ~/.claude/scripts/task_manager.py

Input (from /execute-transition + prior outputs):

• $TICKET
• $WORKTREE_PATH
• $PROMPT_PATH, $PRD_PATH, $QUALITY_PATH
• $RESUME_MODE
• $TEST_FILES_CREATED — tests created in TDD-RED
• $E2E_DOMAIN_PATH — <WORKTREE_PATH>/e2e/tests/<dominio>
• $E2E_TARGET_SPECS — targeted E2E specs created in TDD-RED

Step 6 — CAIO_IMPLEMENTATION_LOOP

Dispatch Caio to implement the feature with strict looping:

1. Run targeted tests (unit + e2e from TDD-RED)
2. Analyze failures
3. Fix production code
4. Re-run tests
5. Repeat until GREEN (tests_failing = 0)

Critical rule: never modify test files unless user explicitly authorizes.

Context:

ticket:              <TICKET>
project_root:        <WORKTREE_PATH>
prompt_path:         <PROMPT_PATH>
prd_path:            <PRD_PATH>
quality_path:        <QUALITY_PATH>
agents_md_path:      <WORKTREE_PATH>/AGENTS.md
e2e_docs_path:       <WORKTREE_PATH>/e2e/docs/
e2e_domain_path:     <WORKTREE_PATH>/e2e/tests/<dominio>
e2e_target_specs:    [<WORKTREE_PATH>/e2e/tests/<dominio>/<flow>.spec.ts]
test_files_created:  [<paths>]
resume_mode:         "continue" | "restart"

Dispatch instruction:

"Você é o Caio. Execute a implementação em loop GREEN para <TICKET>.

Regras obrigatórias:

• Leia PRD, PROMPT, QUALITY, <WORKTREE_PATH>/AGENTS.md e <WORKTREE_PATH>/e2e/docs/.
• Faça loop: testar -> corrigir código de produção -> retestar, até ficar GREEN.
• Execute E2E somente via <WORKTREE_PATH>/e2e/tests/<dominio> (e2e_target_specs).
• NUNCA altere testes sem autorização explícita do usuário.
• Se detectar que alteração de teste é necessária, pare e solicite autorização.
• Não apague/reescreva código sem conflito explícito com PROMPT.
• Protocolo djalma é proibido.

Retorne JSON:

{
  \"ticket\": \"<TICKET>\",
  \"implementation_passed\": true/false,
  \"tests_status\": \"all_passing (GREEN) | failing | blocked_waiting_user_auth\",
  \"tests_passing\": <number>,
  \"tests_failing\": <number>,
  \"loops_executed\": <number>,
  \"test_change_requested\": true/false,
  \"test_change_reason\": \"<when requested>\",
  \"execution_summary\": {
    \"scope\": \"frontend | backend | fullstack\",
    \"agents_dispatched\": [],
    \"code_review_rounds\": <number>,
    \"ux_review\": \"approved | skipped\",
    \"qa_verdict\": \"approved | rejected\"
  },
  \"ready_for_code_review\": true/false
}
```"

Wait for Caio result in foreground.

✅ SUCCESS PATH

Proceed only when:

• implementation_passed = true
• tests_status = all_passing (GREEN)
• tests_failing = 0
• ready_for_code_review = true

Then:

{
  "ticket": "<TICKET>",
  "implementation_passed": true,
  "tests_status": "all_passing (GREEN)",
  "ready_for_code_review": true
}

Proceed to next phase: /execute-code-review

⛔ BLOCKED/FAILURE PATH

If tests_status = blocked_waiting_user_auth or test_change_requested = true:

• Stop and ask user authorization explicitly before any test modifications.

If tests remain failing:

• Stop and return failure details for user decision.

Safety rules

• Foreground-only.
• Never alter tests without explicit user authorization.
• Always reference <WORKTREE_PATH>/AGENTS.md and <WORKTREE_PATH>/e2e/docs/ as source of truth.
• Run E2E from <WORKTREE_PATH>/e2e/tests/<dominio> targets only.