Arcgentic: plan, build, and audit gates for coding agents
Gated plan → development → self-audit → external audit workflow for Codex and Claude Code, so AI-written code only counts as done with evidence.
Free
Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsArcgentic: plan, build, and audit gates for coding agents
Example session with this skill installed
I need to refactor a complex React project but I want a strict process with planning and auditing. How do I get started with Arcgentic?
- Read your context and instructions
- Compiled the arcgentic
- Generated the document
I will guide you through setting up Arcgentic, a harness-engineering layer that adds roles, handoffs, and audit gates to your development workflow. This ensures your React refactor follows a formal Plan -> Develop -> Audit cycle.
Arcgentic Installation & Workflow
To begin using high-integrity rounds, you need to install the Arcgentic plugin or CLI depending on your current environment.
Setup Commands
For Claude Code:
/plugin marketplace add Arch1eSUN/Arcgentic
/plugin install arcgentic@arc-studio
For Codex (Global CLI):
npm install -g arcgentic
arcgentic install-codex-local
The Arcgentic Round Structure
Once installed, every task follows this gated lifecycle:
- Planner: Analyzes your React components and creates a scoped handoff.
- Developer: Performs the refactor and writes a self-audit report.
- Test (optional): Runs a realistic user test of the result.
- Auditor: Reviews the code independently. Returns a
PASSverdict or a specific fix list. - Closure: The round only closes when evidence of a
PASSis recorded.
Optional Status Monitoring
If you have Python 3.13+, you can add the MCP status panel to see role dispatch progress in real-time:
claude mcp add --transport stdio arcgentic -- uvx --python 3.13 --from 'arcgentic[mcp]' arcgentic mcp-serve
Next steps
- Run the installation command for your agent environment.
- Inside your project, ask:
Use Arcgentic to build this idea: refactor the /src/components directory. - On the first run, Arcgentic recommends a session mode (faster single-session subagents, or slower dedicated role threads with stronger audit isolation). Confirm or override it.
- Review the first handoff from the Planner before implementation begins.
arcgentic-plan-build-and-audit-gates-for.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Most AI-generated code ships without accountability: you get a block of code, but no proof it was planned, tested, or reviewed by anyone other than the agent that wrote it. Arcgentic is an open-source (MIT) harness for Codex and Claude Code that closes that gap with a gated cycle of planning, development, self-audit, optional user testing, and independent external audit. This listing is the free entry skill: it tells your agent when Arcgentic fits and how to install the full plugin from GitHub, npm, or PyPI.
What it does
- Enforce quality gates by requiring evidence-based pass/fail markers before code is accepted.
- Implement role-based handoffs between Planner, Developer, optional Test, and Auditor roles, either as named subagents in one session or as fixed role threads.
- Point you to the right install path for Claude Code or Codex so you can set up the full open-source harness yourself.
- Support complex refactors by maintaining context through multiple rounds of development and review.
How it works
- Trigger the skill by asking about Arcgentic or requesting a high-integrity coding workflow.
- Verify the environment to check if the full Arcgentic plugin is already installed in your agent.
- Show the install commands for Claude Code or Codex so you can run them yourself; nothing is installed without your go-ahead.
- Execute rounds where a Planner defines scope, a Developer implements code, and an Auditor provides the final verdict.
Frameworks & tools
Built for Claude Code and Codex. The Claude Code install uses the plugin marketplace; the Codex install uses npm (Node.js 18+). The command-line helper and the optional MCP status panel need Python 3.13+. Arcgentic itself is open source (MIT) on GitHub, PyPI, and npm.
Why this beats prompting it yourself
Standard prompts often result in the agent "grading its own homework." This skill sets up a formal harness where the Developer role must satisfy an independent Auditor role, preventing the hall-of-mirrors effect in complex codebases.
Use cases
- Legacy refactoring where you need a plan and audit trail for every change.
- High-stakes features requiring documented testing and independent validation.
- Multi-session builds that need a structured state machine to manage progress.
Known limitations
Full orchestration and role dispatch only work in Codex and Claude Code. Other agents are limited to the Python CLI helper. Python 3.13 is required for MCP features.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scanned
Verified clean 1 day ago
- Free to download with an account
Needs access to
Github
Pypi
Npmjs