# Prd Auto Executor

> Semi-automated PRD execution with Git safety and intelligent MCP selection

- Skill: `dongho5270/prd-auto-executor` (Agent Skill, multi-file: 29 files)
- Install (CLI): `npx skillmds@latest add dongho5270/prd-auto-executor`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dongho5270/prd-auto-executor/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Product & Planning
- Author: DONGHO5270 (https://skillmd.com/u/dongho5270)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dongho5270/prd-auto-executor

---


# prd-auto-executor Skill

Semi-automated PRD checklist execution with Git safety.

## Mission

Execute PRD checklists with:
- Git checkpoint safety (automatic rollback on failure)
- SDD-style verification (test/build/lint)
- User approval at every step
- State persistence for session resume

## Execution Flow

### Phase 1: Initialize
- Load `execution_state.json` (resume if exists)
- Load checklist JSON from prd-implementation-tracker

### Phase 2: Task Loop
```
FOR EACH task:
  1. Create Git checkpoint
  2. Request user approval (승인/건너뛰기/중단)
  3. Execute task (Claude performs work)
  4. Validate (npm test/build/lint)
  5. Success → Git commit | Failure → Rollback
  6. Save state
```

### Phase 3: Complete
- Summary report (completed/failed/skipped)

## Safety Rules

| Rule | Description |
|------|-------------|
| Checkpoint First | Always create checkpoint BEFORE task |
| Rollback on Fail | Rollback on any validation failure |
| Save State | Save after every operation |
| Approval Required | Never auto-execute without approval |

---

**Detailed protocol and examples**: See `prompt.md`