Menu
Full SDLC workflow - from task to PR. Plans, audits adversarially, implements, runs quality gates, and opens a PR.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Generates a context-specific audit brief and executes review via code-reviewer agent. Usable standalone or as a building block within other skills.
Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes — Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/PRD asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".
Generates a PR description from the current branch diff and project conventions. Output is a ready-to-use markdown body.
Configure this repo for the engineering skills — set up its issue tracker, triage label vocabulary, and domain doc layout. Run once before first use of the other engineering skills.
Validates project.yaml is well-formed. Checks required fields, valid values, resolvable paths, and detects typos.
Validates code against all project architecture conventions. Reads project.yaml to determine applicable checks and runs them. Replaces individual check-* skills.
| name | implement |
| description | Full SDLC workflow - from task to PR. Plans, audits adversarially, implements, runs quality gates, and opens a PR. |
| skills | interface-scout, ddd-scaffold, run-and-fix, audit, pr-description |
From task description to merge-ready PR with adversarial audits at every gate.
/implement <task description>
--no-pr — Skip PR creation.--no-audit — Skip adversarial audits for trivial changes.Read project.yaml (default .ai/) for architecture, conventions, quality gates, testing, and git config.
Parse input (ticket URL, issue, description). Identify affected layers.
Scout the codebase:
interface-scout for existing patternsCreate implementation plan, layer-ordered:
Include: file paths, key changes, order, testing strategy, risks.
Present to user for approval.
Invoke audit with directed context:
Show findings. Incorporate blockers + important items into refined plan.
STOP — Wait for explicit user approval before implementing.
After approval, execute without human checkpoints.
New feature module → invoke ddd-scaffold. Extending existing code → skip.
Spawn implementer subagent:
You are implementing the following plan precisely.
## Task
{task}
## Plan
{refined plan}
## Context
{codebase analysis from Step 1}
## Rules
- Follow the plan. Deviate only for obvious plan bugs.
- Production-ready code. No TODOs, no placeholders.
- Follow existing patterns from reference implementations.
- Layer order: ports → domain → adapters → app service → tests → wiring.
- Report (follow `rule://writing-style`): files modified, functions added/changed.
Invoke run-and-fix (format → lint → typecheck → tests). Continue to audit even if a gate fails after max retries.
Invoke audit with directed context:
Expect a PASS/FAIL verdict.
If audit found blockers or important issues, spawn fixer subagent:
Fix the audit findings below.
## Findings
{audit output from Phase 4}
## Rules
- Fix ALL blockers.
- Fix important items unless explicitly not applicable.
- Skip nitpicks unless trivial.
- Run tests after fixes.
- If no blockers/important: confirm and make no changes.
Invoke audit in verification mode:
Expect a PASS/FAIL verdict.
--no-pr)On PASS:
git.branch_formatgit.commit_formatpr-descriptionOn FAIL: report remaining issues. Do not create PR.
project.yaml canonically.