# Plan Implementation

> Disciplined execution of approved plans with step-by-step verification, phase checkpoints, failure investigation, and mandatory code/security reviews.

- Skill: `a5c-ai/plan-implementation` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add a5c-ai/plan-implementation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/a5c-ai/plan-implementation/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: a5c-ai (https://skillmd.com/u/a5c-ai)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/a5c-ai/plan-implementation

---


- For medium and high stakes changes (low stakes can proceed inline)
- When structured execution with verification is needed

## Process

1. **Load plan** - Validate plan exists and is approved
2. **Stakes enforcement** - High: halt without plan. Medium: ask. Low: proceed.
3. **Worktree isolation** - Offer git worktree based on stakes level
4. **Execute steps** - For each: mark in-progress, locate files, read, modify, verify, mark complete
5. **Phase checkpoints** - Summarize and ask: continue, review, or pause
6. **Failure handling** - Stop immediately, investigate, propose fix, get approval for deviations
7. **Code review** - Run code-reviewer agent (APPROVE / APPROVE_WITH_NITS / REQUEST_CHANGES)
8. **Security review** - Run security-reviewer agent (halt if failed)
9. **Completion summary** - Steps completed, files changed, test status, plan location

## Key Rules

- Follow the plan strictly; deviations require explicit approval
- Verify before declaring done; run verification after each step
- Track progress visibly via plan document updates
- Read files before modifying them
- Complete code and security reviews before finishing

## Tool Use

Invoke via babysitter process: `methodologies/rpikit/rpikit-implement`