--- slug: "ancplua-claude-plugins" source_type: "readme" source_url: "https://cdn.jsdelivr.net/gh/ANcpLua/ancplua-claude-plugins@main/README.md" repo: "https://github.com/ANcpLua/ancplua-claude-plugins" source_file: "README.md" branch: "main" --- # ancplua-claude-plugins Claude Code plugin marketplace — an agent-teams-aware advanced software-engineering base set. 14 plugins for parallel agent orchestration, PR-to-merge ferrying, dependency migration, first-principles repo transformation, parallel code review, cognitive guardrails, behavior-first test quality, plugin/skill evaluation, X/Twitter data workflow planning, and code-elegance work. ## What this does Built around Claude Code's multi-agent capabilities — from one-shot subagent fan-out (exodia spawns up to 12 workers) to the experimental agent-teams feature (council reconciles an Opus team against a /deep-research report via the Teams API). Each phase is gated — work only advances when the gate passes. No manual babysitting. ## Plugins 14 plugins, 24 commands, 24 skills, 30 agents: | Plugin | What it does in plain language | |-----------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | **exodia** | Summons up to 12 AI workers that tackle different parts of your code simultaneously. One finds problems, another fixes them, another reviews the fix — all at the same time | | **council** | Council for complex tasks. Opus captain decomposes and runs the /deep-research dynamic workflow, an Opus synthesizer reasons over it, an Opus clarity agent gap-checks it live, and an Opus janitor flags bloat (Teams API) | | **elegance-pipeline** | Multi-agent pipeline that scores code elegance and refactors the weakest files through gated stages — plus a standalone simplifier for quick diff cleanups | | **metacognitive-guard** | Watches the AI while it works. If it's about to cut corners, guess instead of verify, or say "done" when it isn't — this blocks it before the mistake happens | | **feature-dev** | A guided process for building new things: understand what exists, design the plan, build it, review it. No skipping steps | | **mutation-minded-testing** | Judges whether your tests actually catch bugs, not just cover lines. Four agents review architecture, test quality, and expressiveness | | **cc-plugin-eval** | Evaluates and scores Claude Code plugins and skills: static analysis, token budgeting, and benchmarking across every component type | | **html-effectiveness** | Routes output into a single self-contained HTML file — dashboards, reports, code reviews, slide decks — instead of a wall of markdown | | **nuget-opensrc** | Fetches the exact source a NuGet package was built from, so the agent greps real code instead of guessing API shapes | | **charon** | Babysits a pull request all the way to merge — fixes the CI, repairs conflicts, checks reviewer suggestions against real code, and never just says "wait". Say "merge my PR" and it ferries it home | | **derot** | Hunts truth-drift — stale comments, dead docs, wrong versions, dependency rot — and proposes fixes; `/depmigrate` migrates a package to its newer API and deletes the old code, gated by an adversarial refutation check | | **nihil** | The last-resort cleanup judge: decides whether each artifact still earns its existence and plans the smallest transformation that deserves to exist — nothing is sacred, and every deletion must survive a "prove it's not safe to remove" check | | **tomevault-publish** | Publishes a skill, config, or plugin to TomeVault as a high-grade Tome, and explains the Skill / Tome / AGENTS.md model and grading rubric behind it | | **x-twitter-scraper** | Plans Xquik REST API, MCP, SDK, webhook, export, monitor, and confirmation-gated workflows for X/Twitter data | | **heimdall** | Tells you which model *actually* served each turn — read from the session transcript, not guessed — and shouts when Claude Code silently falls back from Opus to Fable. Answers "am I still on Opus?" with evidence, plus provider/region/auth routing context | ### How does this work without failing? Every step is a gate. Work only moves forward if the gate says PROCEED. If it says HALT — the work stops, gets diagnosed, and gets fixed before anything else continues. No hoping. No skipping. No "it probably works." ## Install Add the marketplace, then install plugins individually: ```bash # Add the marketplace /plugin marketplace add ANcpLua/ancplua-claude-plugins # Install plugins you want /plugin install exodia@ancplua-claude-plugins /plugin install council@ancplua-claude-plugins /plugin install metacognitive-guard@ancplua-claude-plugins /plugin install elegance-pipeline@ancplua-claude-plugins /plugin install cc-plugin-eval@ancplua-claude-plugins /plugin install x-twitter-scraper@ancplua-claude-plugins ``` ## Technical details Claude package counts: 17 plugins, 25 commands, 34 skills, 31 agents. Claude Code only — no other CLI mirrors, no non-Claude models. (Counts cover shipped marketplace plugins; cc-plugin-eval's `fixtures/` test corpus is excluded.) Tri-AI review system: Claude, Copilot, and CodeRabbit review PRs independently. ## GitHub review automation Two workflows drive PR automation: - **`.github/workflows/auto-merge.yml`** — enables native GitHub auto-merge (squash) for Claude-authored PRs (`claude/` branches) from this repo. Renovate enables its own native auto-merge, so it needs no entry here. - **`.github/workflows/claude.yml`** — runs the Claude Code action on `@claude` mentions in issues, pull-request reviews, and comments. ```text plugins/ ├── exodia/ # parallel agent orchestration (7 commands + 2 skills) ├── council/ # Opus council + /deep-research workflow (Teams API) ├── elegance-pipeline/ # code elegance scoring + gated refactoring + standalone simplifier ├── metacognitive-guard/ # quality gates + commit integrity + CI verification ├── feature-dev/ # guided feature development + code review ├── mutation-minded-testing/ # behavior-first test quality (4 agents) ├── cc-plugin-eval/ # plugin & skill evaluator (static analysis + benchmarking) ├── html-effectiveness/ # route output to a self-contained HTML file ├── nuget-opensrc/ # commit-pinned NuGet → source fetcher ├── charon/ # ferries a GitHub PR to merge — never waits forever ├── derot/ # truth-drift hunt + dependency migration (/depmigrate) ├── nihil/ # evidence-gated first-principles repo transformation ├── tomevault-publish/ # publish skills / configs / plugins to TomeVault as Tomes └── x-twitter-scraper/ # Xquik X/Twitter data workflow planning ``` ## Links - [CHANGELOG](https://github.com/ANcpLua/ancplua-claude-plugins/blob/HEAD/CHANGELOG.md) - [Architecture](https://github.com/ANcpLua/ancplua-claude-plugins/blob/HEAD/docs/ARCHITECTURE.md) - [Claude Code Plugins docs](https://code.claude.com/docs/en/plugins) ## License [MIT](https://github.com/ANcpLua/ancplua-claude-plugins/tree/HEAD/LICENSE)