By Xuoon
Sync, Audit, Prune & Init für Agent-Doku (CLAUDE.md/AGENTS.md + Rules-Verzeichnisse). Approval-gated, evidenz-basiert, löschen-bevorzugt.
Comprehensive scored review of all agent docs (CLAUDE.md/AGENTS.md plus rules directories) with bidirectional claim verification, coverage sweep, paths-glob validation and prune-sweep. Manual-only — invoke via /agent-docs:audit. Optional args — subtree path; `solo` runs without subagents; `quick` skips scoring and coverage (red flags + below-B list only).
Bootstrap agent docs (CLAUDE.md + rules structure) in a repo or subtree that has none yet. Use ONLY when the user explicitly asks to initialize/set up agent docs — "init claude docs", "setup CLAUDE.md", "bootstrap agent docs", "leg mir eine CLAUDE.md an". Produces minimal, code-verified docs (no boilerplate, no empty scaffolds), approval-gated. Optional arg — subtree path for monorepo packages.
Dedup and deletion pass over agent docs (CLAUDE.md/AGENTS.md plus rules directories). Use ONLY when the user explicitly asks to declutter, dedupe or shrink docs — "was kann weg", "nichts doppelt dokumentieren", "docs entschlacken", "prune docs" — or as a counter-check right after an audit that added content. Never trigger this just because docs look long. Approval-gated, file-line evidence, deletion-first. Optional arg — subtree path.
Surgical post-session sync of agent docs (CLAUDE.md/AGENTS.md plus rules directories like `.claude/rules/**`), driven by the actual code diff. Use whenever code was touched in this session or branch, even if the user doesn't mention docs — and on "sync docs", "doku aktualisieren", "passen die rules noch?", "update CLAUDE.md". Approval-gated, file-line evidence, prefers deletion, never rewrites whole files. Optional args — base branch (diff vs merge-base) and/or subtree path.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Persönlicher Plugin-Marketplace für Claude Code. Philosophie aller Plugins: Evidenz statt Vermutung, Vorschlag vor Edit (Approval-Gates), löschen bevorzugt, concise Output.
| Plugin | Befehle | Zweck | Status |
|---|---|---|---|
agent-docs 1.0.0 | /agent-docs:sync :audit :prune :init | Agent-Doku (CLAUDE.md/AGENTS.md + Rules) mit der Code-Realität konsistent halten | stabil |
git-work 0.1.0 | /git-work:commit :changelog | Logisch geschnittene Commits + concise Changelogs im Haus-Stil | neu |
deps 0.1.0 | /deps:bump | Alle Dependencies aufs letzte offizielle stabile Release (bun-first) | neu |
dead-code 0.1.0 | /dead-code | Toten Code finden, belegen, nach Freigabe löschen | neu |
Geplant (noch nicht gebaut): deps:audit (CVEs/EOL/Lizenzen), compose-maintain (Docker-Hygiene), runbook (Ops-Doku).
Auf jedem Gerät einmalig in Claude Code:
/plugin marketplace add DEIN-GITHUB-USER/claude-plugins
/plugin install agent-docs@labi
/plugin install git-work@labi
/plugin install deps@labi
/plugin install dead-code@labi
Danach Claude Code neu starten. Updates kommen über /plugin update (bzw. Auto-Update), gesteuert über das version-Feld der jeweiligen plugin.json.
Pro Plugin: Semver in .claude-plugin/plugin.json + Eintrag im CHANGELOG.md des Plugin-Ordners. Faustregel: Befehls-/Argument-Änderungen = minor, reine Instruktions-Verbesserungen = patch, Umbenennungen/Entfernungen = major. (/git-work:changelog plugins/<name> pflegt die Changelogs selbst.)
Einträge in .claude-plugin/marketplace.json können statt auf lokale Ordner direkt auf fremde GitHub-Repos zeigen — die Dateien liegen dann nie in diesem Repo, /plugin update zieht vom Original:
{
"name": "fremdes-plugin",
"source": { "source": "github", "repo": "owner/repo", "ref": "v1.2.0" }
}
ref (Tag/Commit) pinnen statt HEAD folgen — fremder Code mit Hooks/MCP-Servern läuft lokal, Vertrauen + Pinning sind Pflicht.
python scripts/validate.py prüft marketplace.json, alle plugin.json, SKILL.md-Frontmatter, Referenz-Pfade und Markdown-Links. Läuft automatisch bei jedem Push (GitHub Action). Lokal: pip install pyyaml einmalig.
claude-plugins/
├── .claude-plugin/marketplace.json # der Katalog (name: "labi")
├── plugins/
│ ├── agent-docs/ # Multi-Skill: skills/<name>/SKILL.md → /agent-docs:<name>
│ ├── git-work/
│ ├── deps/
│ └── dead-code/ # Single-Skill: SKILL.md im Root → /dead-code
├── scripts/validate.py
├── .github/workflows/validate.yml
└── CONTRIBUTING.md # neues Plugin in 5 Schritten
npx claudepluginhub xuoon/skills --plugin agent-docsDependency-Pflege, bun-first. Aktuell: /deps:bump — alles auf das letzte offizielle stabile Release, mit Plan, Major-Risiko-Notizen und Verify.
Git-Workflow im Haus-Stil: logisch geschnittene Commits und concise Changelogs. Approval-gated.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Frontend design skill for UI/UX implementation
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Memory compression system for Claude Code - persist context across sessions
Marketing skills for AI agents — conversion optimization, copywriting, SEO, paid ads, ad creative, and growth
Standalone image generation plugin using Nano Banana MCP server. Generates and edits images, icons, diagrams, patterns, and visual assets via Gemini image models. No Gemini CLI dependency required.