Arcgentic plugin
A plugin for Codex and Claude Code that runs a task through a chain of roles from planning to an independent audit
- 285 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Dispatches agent roles through its own threads and state in .agentic-rounds, which it reads and writes without asking at each step
- The hooks fallback adds entries to .claude/settings.local.json
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/arcgenticDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Dispatches agent roles through its own threads and state in .agentic-rounds, which it reads and writes without asking at each step
- The hooks fallback adds entries to .claude/settings.local.json
This is third-party code. Review the repository files before installing.
What it does
Arcgentic wraps a Codex or Claude Code session in a chain of roles. Orchestrator sets the task, Planner writes the plan, Developer implements it and checks its own work. Then Auditor checks the result independently and returns PASS, NEEDS_FIX or AUDIT_INCOMPLETE. An optional realistic test can run between development and audit. The current chat session becomes the orchestrator. It creates role threads or reuses existing ones, sends each the right prompt and waits for the return signal, so you never switch threads by hand. In Codex the orchestrator creates and names the role threads itself. In Claude Code the roles run through a broker on the built-in subagent tools. For a new idea the orchestrator proposes a project mode, fast or with stronger review, and asks you to confirm or change it before planning starts.
Who it is for. For heavy Codex and Claude Code users who need a repeatable plan, build and audit chain on complex tasks.
Good fit when
- Multi-round Codex development where Planner, Developer and Auditor must be distinct roles with their own prompts and stop conditions
- You want an independent audit: the PASS or NEEDS_FIX verdict comes from a role that did not write the code
- The model keeps declaring tasks done although nobody ran the tests
- Work runs over several rounds and task state must persist between them
Not a fit when
- You need Claude Code work that switches across several sessions: the developer has not verified that mode
- You need a five-minute edit: the plan, build and audit chain is built for complex multi-round work
- You want a standalone AI agent: Arcgentic is a layer around Codex and Claude Code with no agent of its own
Example request
Use Arcgentic to build this idea: a small expense tracker with categories and a monthly reportLimitations
In Codex the path is fully verified. In Claude Code only the single-session-subagent mode, with a broker on the built-in Agent, SendMessage and ListAgents tools, has passed real testing. The developer calls the multi-session-subthread mode and the hooks install unverified, and hooks are only for sessions without those tools. The Python CLI is published separately on PyPI, and the npm package holds only plugin assets and a Codex install helper. Arcgentic does not replace your own tests and review; it only makes skipping them more visible.
How to disable. In Claude Code remove the plugin with `/plugin uninstall arcgentic@arc-studio` and the marketplace with `/plugin marketplace remove arc-studio`. If you installed hooks, remove the added entries from .claude/settings.local.json and delete the .agentic-rounds state folder.
FAQ
How is Arcgentic different from normal planning in Claude Code?
Arcgentic creates separate threads for the Planner, Developer and Auditor roles with their own prompts and stop conditions. A task closes only after a PASS verdict from an auditor that did not write the code.
Does Arcgentic work the same way on Codex and Claude Code?
No. In Codex role dispatch was verified on a real project, and the orchestrator switches threads itself. In Claude Code only one broker mode is verified, and the developer calls the other paths unverified.
Do I need to install hooks for Claude Code?
Usually not: if the session has the built-in subagent tools, hooks are not needed. The install-hooks command stays a fallback for sessions without them.
Where can I install Arcgentic from?
The package with the Codex install helper is on npm, the Python CLI with checks and version state is on PyPI, and Claude Code gets a plugin marketplace manifest. Sources and releases are on GitHub.
How does Orchestrator pick a project mode for a new idea?
Before planning the orchestrator assesses the idea and proposes the fast mode for a small project or the stronger-review mode for a large one. The user confirms or changes the choice.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code and Codex plugin with skills for code research, tests, refactoring and independent review of finished changes
A large Claude Code and Cursor plugin: development skills, specialist agents and hooks with quality gates and git protection
A Claude Code plugin with seven specialized subagents and approval gates: the agent proposes a plan and waits for confirmation before touching code
A Grid Dynamics plugin that loads a team's shared engineering standards, architecture and workflows into a coding agent session
OpenAI's official plugin: Codex reviews of your changes and task delegation to Codex from inside a Claude Code session
A governance layer for agentic coding: enforced gates for tests, reviews, security and decisions, shared across Claude Code, Codex CLI and Pi
README in short
The README tells how Arcgentic grew out of the development discipline of the Moirai project and tabulates verification status by platform and distribution channels. Three install routes and a short example of work without and with Arcgentic follow.