Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
72 lines (57 loc) · 3.26 KB

CONTRIBUTING.md

File metadata and controls

72 lines (57 loc) · 3.26 KB
Raw

Contributing to WIGTN Plugins

Thanks for your interest! This repo is a Claude Code plugin marketplace — one unified plugin (wigtn-plugins) made of agents, commands, and skills. The notes below keep contributions consistent and mergeable.

한국어 사용자를 위한 콘텐츠는 한국어로 작성하되, 코드·커밋·이 문서는 영어를 유지합니다.

Repo layout

.claude-plugin/marketplace.json   # marketplace entry (name, version, plugin list)
plugins/wigtn-plugins/
├── .claude-plugin/plugin.json     # plugin manifest (agents/commands/skills arrays)
├── agents/      *.md              # one file per agent
├── commands/    *.md              # one file per slash command
├── skills/      <name>/SKILL.md   # one folder per skill
├── scripts/     *.sh              # bundled scripts commands call directly
└── hooks/hooks.json

Adding an agent / command / skill

  1. Create the file/folder under the matching directory.
  2. Register it in plugins/wigtn-plugins/.claude-plugin/plugin.json (add to the agents / commands / skills array). A skill that is not in the array is not exposed by the plugin.
  3. Update the counts everywhere — this is enforced by CI (see below):
    • plugins/wigtn-plugins/.claude-plugin/plugin.json description (N agents, …)
    • .claude-plugin/plugin.json (root manifest description)
    • .claude-plugin/marketplace.json (both descriptions)
    • README.md, README.ko.md, README.cn.md (the N-Skills / N-Agents badges and the skill/agent tables)
    • CLAUDE.md architecture block
  4. User-facing content (skills, commands) is written in Korean; keep technical terms in English.

Conventions

  • Commits: Conventional Commits — feat(skill):, fix:, docs:, chore:, …
  • Branches: feat/<name>, fix/<desc>, refactor/<desc>.
  • Skill frontmatter: name, description (with trigger keywords), optional allowed-tools. See existing skills for the pattern.
  • Mermaid in generated docs: quote every label (["..."]), use only valid shapes, no mindmap (use flowchart LR).
  • Bundled scripts: a command that needs several read-only lookups (git, gh, file probes) calls one script under scripts/ via bash "${CLAUDE_PLUGIN_ROOT}/scripts/<name>.sh" instead of listing the commands for the model to run one by one. Return a single JSON object, cap unbounded fields, and never let a missing tool abort the script. Keep the decision in the command (the model), not in the script. Add a line telling the model not to run the individual commands anyway — without it the model re-runs them to confirm.

Versioning & releases

Single source of truth: the version in marketplace.json and plugin.json must match, and README / CLAUDE.md must state the same number. Bump with SemVer: fix → patch, feat → minor.

Before you open a PR

Run the same check CI runs:

python3 .github/scripts/validate_plugin.py

It verifies that JSON is valid and that the stated counts, the plugin.json arrays, and the files on disk all agree — and that the version is consistent. PRs that drift will fail this check.

Fill in the PR template, link any related issue, and keep one concern per PR.