Argus plugin
Argus runs long engineering and research tasks with four roles, where a reviewer reads results but edits nothing.
- 146 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Long autonomous campaigns executing code and shell commands; the only boundary is stops for credentials, payment and publication.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/argusDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Long autonomous campaigns executing code and shell commands; the only boundary is stops for credentials, payment and publication.
This is third-party code. Review the repository files before installing.
What it does
Argus splits work into Manager, Planner, Engineer and Reviewer roles. The manager moves a campaign between stages, the planner sets the next task and the evidence it needs, the engineer does the work, and the reviewer issues a verdict read-only and may return `blocked`. The worker cannot declare its own work complete. Credentials, payment, irreversible actions and publication always stop for a human. GitHub Copilot CLI, OpenAI Codex CLI, Claude Code, Cursor CLI, OpenCode and others can serve as the backend. Codex and Claude Code get a plugin: an MCP bridge and host-specific skills, with requests such as "show Argus project status". Approved skills and notes accumulate at project, vertical and global levels.
Who it is for. Researchers and engineers whose tasks run for hours or days and who cannot babysit an agent.
Good fit when
- An experiment or code migration runs for days and needs a process that checks results itself.
- You do not trust an agent to grade its own work and want an independent read-only review.
- You want to start a task from Claude Code or Codex and later check campaign status.
- You want lessons from past campaigns kept as skills and notes instead of lost with the session.
Not a fit when
- You need a quick one-file edit; four roles waste time on it.
- Running install scripts from the web is not allowed; the plugin installs with a `curl ... | sh` command.
- Agents must not get long autonomous runs in your environment.
Example request
Use Argus for this project: start a campaign from the plan, then show the status and the reviewer verdict.Limitations
Argus installs from GitHub source, not PyPI, and `main` differs from a published desktop release. In the docs, the plugin install script lives in the preview repository lbx154/Argus. Argus-Pi is currently a source preview rather than a published package and needs Node.js 22.19 or newer. Trial mode isolates mission records but is not a filesystem sandbox. The medical vertical ships separately as `argus-verticals` and gives no diagnoses.
How to disable. Remove the plugin from Claude Code or Codex, stop campaigns with the `argus` command and uninstall the Python package. Mission data, skills and notes live in the Argus data directory and must be deleted by hand.
FAQ
Which roles exist and what is each barred from?
The manager does not perform the work it admits, the planner does not advance the campaign, the engineer does not declare work done, and the reviewer edits nothing and may return `blocked`.
How do I install the plugin for Claude Code?
Node.js 22.12 or newer is required. The command `curl -fsSL https://raw.githubusercontent.com/lbx154/Argus/main/plugins/argus/install.sh | sh -s -- claude` installs it for Claude Code; use `codex` or `all` instead of `claude`.
What if I would rather not install by hand?
The README has a ready prompt: send it to an installed agent and it follows `docs/agent-install.md` and verifies with `argus doctor`. No passwords or keys go into the chat.
How do the repositories differ?
`microsoft/ArgusAgent` is the official source and `lbx154/Argus` is the development preview. Do not switch channels when updating.
Which agents can serve as the backend?
GitHub Copilot CLI, OpenAI Codex CLI, Claude Code, Cursor CLI, Pi, OpenCode, Grok Build, Qoder and DeepSeek Harness. Prefer the one you already use.
Discussion
No questions or reviews yet.
Related
PluginsA harness around a coding agent: supervisor, reviewer, adversary and local log distiller, each on its own model
SuperARIS builds a persistent research team from Codex, Claude Code and Antigravity: a main agent, an engineer and a writer.
One agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
A team of 14 roles, 92 specialists and a skill library that installs into Claude Code, Codex and other agents via npx.
A layer over coding agents that assembles a temporary specialist team per task inside Claude Code, Codex, Cursor and Gemini
README in short
Nine backends, including Claude Code, Codex and Cursor CLI. The project site and technical report describe campaign results.