Skip to content

CLAUDE.md

Latest commit

 

History

History
69 lines (52 loc) · 3.37 KB

CLAUDE.md

File metadata and controls

69 lines (52 loc) · 3.37 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Repository Overview

This is a Claude Code Plugin (codewright) — a collection of multi-agent skills for Claude Code that automate complex development tasks. Each skill is a self-contained directory under skills/ with a SKILL.md entry point, agent definitions, and reference templates.

Plugin Structure

.claude-plugin/
  plugin.json               # Plugin manifest (name, version, author)
references/                   # Shared references across all skills
  agent-invocation.md         # How to start and communicate with agents
  finding-format.md           # Unified finding format for all analysis agents
skills/
  skill-name/
    SKILL.md                # Main skill definition (frontmatter + workflow)
    agents/                 # Subagent prompt definitions (.md files)
    references/             # Templates, formats, conventions (.md files)

Installation: claude plugin add or test locally with claude --plugin-dir .

Skills

Skill Invocation Agent Pattern
audit-project /codewright:audit-project 5 parallel Explore agents → GitHub Issue creation
codebase-doctor /codewright:codebase-doctor 3-wave: 7 Explore → Fix → Review agents
github-issue-fixer /codewright:github-issue-fixer Sequential: Explore → Plan → Fix → Verify → Commit
refactor-orchestrator /codewright:refactor-orchestrator Teamleader + Scout/Architect/Worker/Test agents (agents/ + references/)
auto-dev /codewright:auto-dev Teamleader + Analyst/Planner/Workers/Reviewers/Fixers (6 phases, review-fix loop)

Skill File Format

SKILL.md files use YAML frontmatter:

---
name: skill-name
description: >
  Multi-line description used for skill matching/triggering.
  Include German and English trigger phrases.
disable-model-invocation: true  # Optional: prevents direct model invocation
---

Key Patterns

  • Agent types: Skills use subagent_type: "Explore" for read-only analysis and general-purpose agents with mode: "auto" for code changes.
  • Inter-agent communication: Agents return structured Markdown responses; no /tmp/ files are used.
  • File conflict avoidance: When multiple agents modify code in parallel, files are strictly partitioned so no two agents edit the same file.
  • Finding format: Both audit-project and codebase-doctor share a standardized finding format defined in their respective references/finding-format.md.
  • User confirmation gates: All skills pause before destructive actions (creating issues, committing, merging) and ask the user.

Versioning & Release

  • Auto-increment patch version on every change (e.g., 0.3.0 → 0.3.1 → 0.3.2). Only bump minor/major when the user explicitly says so.
  • Version must be updated in ALL three files simultaneously:
    • .claude-plugin/plugin.json
    • .claude-plugin/marketplace.json
    • CHANGELOG.md
  • The plugin marketplace reads the version from plugin.json (not git tags). Users must run /plugin update or clear their cache to get the new version.

Language

All skill content, agent prompts, reports, and user-facing output are in English. SKILL.md description fields include both English (primary) and German (secondary) trigger phrases for backward compatibility.