원클릭으로
subagent-driven-development
Use when executing implementation plans with independent tasks in the current session
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Use when executing implementation plans with independent tasks in the current session
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when implementation is complete, all tests pass, and you need to decide how to integrate the work. In this repository, package or lane branches must close through package-first evidence and integration handoff rather than the generic local merge or PR menu.
Use when executing repo-wide or lane-scoped SDK changes in this repository that must follow bounded change packages, qiniu-ai worktree/package commands, evidence bundles, review packets, promotion decisions, and integration-branch handoff. Trigger on requests to open a phase package, spawn or integrate a lane, prepare a review handoff, or land multi-subsystem SDK work without widening scope.
Four-phase debugging framework that ensures root cause investigation before attempting fixes. Never jump to solutions.
Use when starting feature work that needs isolation from the current workspace or before executing implementation plans. In this repository, prefer qiniu-ai worktree commands for lane or package-first delivery; fall back to generic git worktree flows only for non-package branches.
Run verification commands and confirm output before claiming success
Use when you have a spec or requirements for a multi-step task, before touching code. In this repository, multi-subsystem or lane-scoped work must create the package brief first and keep the plan aligned with the bounded package contract.
| name | subagent-driven-development |
| description | Use when executing implementation plans with independent tasks in the current session |
Execute plan by working through tasks sequentially, with two-stage self-review after each: spec compliance review first, then code quality review.
Core principle: One task at a time + two-stage self-review (spec then quality) = high quality, focused iteration
Use when:
vs. Executing Plans (batch mode):
┌─────────────────────────────────────────────────────────┐
│ 1. Read plan, extract all tasks, create task checklist │
└──────────────────────────┬──────────────────────────────┘
▼
┌─────────────────────────────────────────────────────────┐
│ Per Task: │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ 2. Implement task following plan steps exactly │ │
│ └───────────────────────────┬─────────────────────────┘ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ 3. Self-Review: Spec Compliance │ │
│ │ - Does code match what plan specified? │ │
│ │ - All steps completed? │ │
│ │ - Tests written and passing? │ │
│ └───────────────────────────┬─────────────────────────┘ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ 4. Self-Review: Code Quality │ │
│ │ - Clean, readable code? │ │
│ │ - No obvious bugs or edge cases missed? │ │
│ │ - Follows project conventions? │ │
│ └───────────────────────────┬─────────────────────────┘ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ 5. Commit and mark task complete │ │
│ └─────────────────────────────────────────────────────┘ │
└──────────────────────────┬──────────────────────────────┘
▼
┌─────────────────────────────────────────────────────────┐
│ 6. More tasks? → Loop back to step 2 │
│ All done? → Final review of entire implementation │
└──────────────────────────┬──────────────────────────────┘
▼
┌─────────────────────────────────────────────────────────┐
│ 7. Use finishing-a-development-branch skill │
└─────────────────────────────────────────────────────────┘
Ask yourself:
If any NO: Fix before proceeding.
Ask yourself:
If any NO: Fix before proceeding.
## Task 1: Add User Authentication
[Read task from plan]
[Implement following plan steps exactly]
[Run: npm test -- auth.test.ts]
### Spec Compliance Self-Review
✓ JWT middleware added as specified
✓ Login endpoint created with correct path
✓ Tests written for happy path and error cases
✓ All tests passing
### Code Quality Self-Review
✓ Clean separation of concerns
✓ Error messages are user-friendly
✓ Follows existing auth patterns in codebase
[Commit: feat(auth): add user authentication]
[Mark task complete]
## Task 2: Add User Profile
[Continue to next task...]
After all tasks complete:
Then use finishing-a-development-branch skill.
When blocked: Stop and ask for clarification rather than guessing.