Skip to main content

plan-implementation

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

Quellinformationen

Repository
a5c-ai/babysitter
Letzte Quellaktivität
1. Juni 2026 um 07:46
Erkannte Sprache von SKILL.md
Englisch
Sterne
1.832
Forks
113

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
2 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
plan-implementation
description
Disciplined execution of approved plans with step-by-step verification, phase checkpoints, failure investigation, and mandatory code/security reviews.
allowed-tools
Read, Write, Edit, Bash, Grep, Glob, WebFetch, WebSearch, Agent, AskUserQuestion
graph
{"domains":["domain:software-engineering"],"skillAreas":["skill-area:agentic-loops","skill-area:orchestration-loop"],"workflows":["workflow:feature-development"],"topics":["topic:developer-experience"],"roles":["role:tech-lead","role:backend-engineer"]}
- 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`
Auf GitHub ansehen