القائمة
Use when an approved plan has independent tasks ready for delegated execution with fresh subagents and two-stage review.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | codex-subagent-execution |
| description | Use when an approved plan has independent tasks ready for delegated execution with fresh subagents and two-stage review. |
| load_priority | on-demand |
Dispatch fresh subagent per task → implementer works → spec reviewer verifies requirements → code quality reviewer verifies quality → fix loops until approved → next task. Use $sdd or $dispatch to activate.
$codex-subagent-execution, $sdd, or $dispatch.codex-plan-writer generates plan and user approves.codex-workflow-autopilot routes to implement phase with independent tasks.$review-feedback or when receiving code review feedback that needs technical evaluation before changes.Announce at start: "I'm using codex-subagent-execution to implement this plan task-by-task."
FRESH SUBAGENT PER TASK + TWO-STAGE REVIEW = HIGH QUALITY
Why subagents: You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused. They should never inherit your session's context or history — you construct exactly what they need. This preserves your own context for coordination work.
Have implementation plan?
|- Yes → Tasks mostly independent?
| |- Yes → Use codex-subagent-execution ($sdd)
| `- No (tightly coupled) → Manual execution
`- No → Write plan first ($plan)
Read plan → Extract all tasks → Create checklist
For each task:
1. Dispatch implementer subagent (agents/implementer-prompt.md)
|- Questions? → Answer, re-dispatch
`- Implements → tests → commits → self-reviews
2. Dispatch spec reviewer (agents/spec-reviewer-prompt.md)
|- Issues? → Implementer fixes → re-review
`- ✅ Spec compliant
3. Dispatch code quality reviewer (agents/code-quality-reviewer-prompt.md)
|- Issues? → Implementer fixes → re-review
`- ✅ Quality approved
4. Mark task complete
All tasks done → Final code review → Finish ($finish or $commit)
Use the least powerful model that can handle each role to conserve cost and increase speed.
| Task Type | Model Tier | Signals |
|---|---|---|
| Mechanical implementation | Fast/cheap | 1-2 files, clear spec, isolated function |
| Integration work | Standard | Multi-file coordination, pattern matching |
| Architecture & review | Most capable | Design judgment, broad codebase understanding |
Parallel implementation subagents are allowed only when all conditions are true:
If write scopes overlap, run tasks sequentially. Parallel speed is not worth merge conflicts or inconsistent architecture.
| Status | Action |
|---|---|
| DONE | Proceed to spec compliance review |
| DONE_WITH_CONCERNS | Read concerns → address if correctness/scope issue → proceed if observation |
| NEEDS_CONTEXT | Provide missing context → re-dispatch |
| BLOCKED | Assess: context → re-dispatch; reasoning → upgrade model; too large → split; plan wrong → escalate to human |
Never ignore an escalation or force retry without changes.
Templates in agents/ directory:
agents/implementer-prompt.md — Dispatch implementer subagentagents/spec-reviewer-prompt.md — Dispatch spec compliance revieweragents/code-quality-reviewer-prompt.md — Dispatch code quality reviewerYou: I'm using codex-subagent-execution to implement this plan.
[Read plan file once]
[Extract all 5 tasks with full text and context]
[Create task checklist]
Task 1: User authentication endpoint
[Dispatch implementer with full task text + context]
Implementer: "Should I use JWT or session tokens?"
You: "JWT — see existing auth middleware in src/middleware/auth.js"
Implementer: Implemented → 5/5 tests pass → committed
[Dispatch spec reviewer]
Spec reviewer: ✅ Spec compliant — all requirements met
[Dispatch code quality reviewer]
Code reviewer: ✅ Approved — clean implementation
[Mark Task 1 complete]
Task 2: Rate limiting middleware
[Dispatch implementer]
Implementer: Implemented → 8/8 tests pass → committed
[Dispatch spec reviewer]
Spec reviewer: ❌ Missing: per-user rate limits (spec says "per-user")
[Implementer fixes]
[Spec reviewer re-reviews]
Spec reviewer: ✅ Compliant now
[Dispatch code quality reviewer]
Code reviewer: Important: Magic number 100 for rate limit
[Implementer extracts RATE_LIMIT_PER_MINUTE constant]
[Code reviewer re-reviews]
Code reviewer: ✅ Approved
[Mark Task 2 complete]
... (remaining tasks)
[Final code review for entire implementation]
[Run $gate for quality verification]
[Use $commit or $finish]
When project uses codex-scrum-subagents:
After all tasks complete:
# Full verification
python codex-execution-quality-gate/scripts/auto_gate.py --mode full --project-root <repo>
# Deploy readiness
python codex-execution-quality-gate/scripts/auto_gate.py --mode deploy --project-root <repo>
Recommended: Set up isolated workspace before starting:
$worktree → create isolated branch → $sdd → execute plan → $finish
codex-plan-writer — Creates the plan this skill executescodex-test-driven-development — Subagents follow TDD for each taskcodex-git-worktrees — Set up isolated workspace before executioncodex-execution-quality-gate — Verify final implementation via $gatecodex-verification-discipline — Evidence before claims, alwayscodex-branch-finisher — After all tasks complete, use $finish for structured completioncodex-scrum-subagents — Team-level coordination overlayagents/implementer-prompt.md: implementer subagent dispatch templateagents/spec-reviewer-prompt.md: spec compliance review templateagents/code-quality-reviewer-prompt.md: code quality review templatereferences/code-review-discipline.md: how to receive and evaluate code review feedback without performative agreementUse for project memory, decisions, handoffs, genome, knowledge index, changelog, and patterns that must persist across sessions.
Use for frontend, backend, mobile, debugging, security, or specialized engineering work that needs focused domain references and strict boundaries.
Use for prototype, MVP, fullstack feature, from-scratch feature, or multi-domain work that needs a SPEC.md before planning or implementation. Do not use for tiny one-file edits.
Use when code changes may require documentation updates; maps git diff to candidate docs without auto-editing documentation.
Use before completion, PR, deploy, handoff, or when verifying code changes. Runs advisory or blocking checks and reports evidence.
Use when the task involves reading, creating, or editing `.docx` documents, especially when formatting or layout fidelity matters; prefer `python-docx` plus the bundled `scripts/render_docx.py` for visual checks.