AGENTS plugin
Claude Code kit with eight phase agents, 21 skills and hooks that keep the agent inside set limits.
- 79 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Installation writes into home directories and with `--hooks` edits `~/.claude/settings.json`.
- The Builder and Committer agents change files and git history and hooks intervene in tool calls.
Install
- Works in
Manual install
git clone https://github.com/mcouthon/agents.gitThen `cd agents` and `./install.sh`: generated files are already committed.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Installation writes into home directories and with `--hooks` edits `~/.claude/settings.json`.
- The Builder and Committer agents change files and git history and hooks intervene in tool calls.
This is third-party code. Review the repository files before installing.
What it does
AGENTS splits Claude Code work into phases: Explorer only reads code and writes a plan, Builder implements, Reviewer checks and Committer prepares the commit. Conductor mode drives a multi-phase task by itself and stops where the owner must decide. The kit holds eight agents, 21 skills and five instruction files by file type. Skills activate by themselves as you talk: `debug` for a failing test, `tech-debt` when hunting code smells, `security-review` for change review, `testing`, `documentation`, `deep-research`. Hooks enforce tool safety. An optional state MCP server gives agents seven tools for task progress and two for the code index.
Who it is for. Claude Code developers who want a controlled process with human checkpoints instead of one agent doing everything.
Good fit when
- Before editing, the code must be researched and the plan approved: Explorer cannot change files.
- A large task runs by phases through Conductor with pauses for approval.
- A failing test is handled by the `debug` skill, which activates from the phrase.
- A pull request review needs a cold Reviewer pass over the whole diff.
Not a fit when
- You need one universal agent without phases and pauses.
- Work happens outside Claude Code: the kit is written for it.
- `~/.claude` must not change: installation copies files into the home directory.
Example request
Use Explorer to study the authorization module and draft a plan, and hand over to Builder only after I approve.Limitations
Installation copies agents, skills and instructions into home directories, for example skills into `~/.claude/skills`, and puts the `a` shell shim in `~/.local/bin`. The `--hooks` flag backs up `~/.claude/settings.json` and adds four hooks, while `--mcp` registers the state server through `claude mcp`. The pipeline assumes a human in the loop: Conductor pauses on owner decisions. The kit is large, and its process can be overkill for a small one-off task.
How to disable. Run `./install.sh uninstall` in the repository clone. Remove the state server entry with `claude mcp remove state-manager`, and hooks are removed by the same script with the `--hooks` flag.
FAQ
How do I start a specific agent?
In Claude Code run `claude --agent AgentName` or write `use AgentName`. Or just talk about the task and skills activate themselves.
How does Conductor differ from the manual flow?
In the manual flow you hand work from Explorer to Builder, Reviewer and Committer yourself. Conductor runs the plan, review, build and commit phases itself and pauses for approval.
Do I need to run `make`?
No, generated files are committed and `./install.sh` works at once. `make` is needed after editing templates.
Where do skills go after installation?
For Claude Code into `~/.claude/skills`. Agents and instructions land in neighboring home directories.
Is the MCP server required?
No, the core workflow runs on prompts alone. The state server is registered with `--mcp` or with `claude mcp add --scope user state-manager`.
Discussion
No questions or reviews yet.
Related
PluginsA ready-made Claude Code setup: agents, slash commands, skills, safety hooks, rules and MCP connections in one install
A set of Claude Code utilities: background skills, expert agents, slash commands and prompt templates for development
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
A universal workflow plugin for Claude Code: 7 agents, 26 commands, 14 skills and 14 hooks for any project
Claude Code Skills and Agents
Claude Code Skills & Agents
A Claude Code plugin bundling skills and sub-agents: deep research, image generation, GitHub work, code review and model switching
README in short
The text lists the kit components, diagrams of the manual and orchestrated workflow, a skills table with trigger phrases and sections on installation, hooks, the state server, model configuration and per-repository push policy.