Codebase Argus CLI
A CLI for maintainers: pull request review, failing CI diagnosis and fork sync analysis built on repository evidence.
- 59 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads patches and logs from GitHub, sends evidence to chosen model providers and handles a GitHub token.
Install
- Works in
Manual install
git clone https://github.com/AaronZ345/codebase-argus.gitThen `npm ci` in the directory and `npm run argus -- review owner/repo#123` for the first run.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads patches and logs from GitHub, sends evidence to chosen model providers and handles a GitHub token.
This is third-party code. Review the repository files before installing.
What it does
Codebase Argus reviews pull requests, failing build logs and long-lived forks from one set of signals: patches, checks, files, branch state, policy gates and local git simulations. The deterministic pass runs locally, read-only and without a model key. Optionally the same evidence goes to one model, several at once or local CLIs such as `codex-cli` and `claude-cli`, and every finding stays tied to a checkable fact. Commands run through `npm run argus`: `review` for a pull request, `ci-log` and `ci-github` for builds and `downstream` for a fork and its upstream. For a fork Argus measures drift, projects conflicts with `git merge-tree` and simulates a rebase in a temporary worktree. Alongside sit an agent playbook `skills/codebase-argus/SKILL.md`, a GitHub Action and a GitHub App.
Who it is for. Maintainers of open source projects and forks for whom one reviewer is not enough and a fully automatic merge bot is too risky.
Good fit when
- A large pull request should go to several models with their findings compared.
- A GitHub Actions build failed and you need the cause and the command to rerun fast.
- A fork trails upstream and you want to see likely conflicts before rebasing.
- Pull requests need an automatic comment with the check outcome in the job summary.
Not a fit when
- You need a bot that merges changes with no human involved.
- The repository is not on GitHub: data sources are built for it.
- You want a replacement for a real reviewer who knows the project architecture.
Example request
Review pull request zhayujie/CowAgent#2965 and show the findings and questions worth asking the author.Limitations
Node.js, npm and git are required, and installation is a clone followed by `npm ci`. Private repositories, higher API limits and GitHub Actions logs need `GITHUB_TOKEN`, while model providers ask for their own `OPENAI_API_KEY` and `ANTHROPIC_API_KEY`. Autofix only builds a branch plan for mechanical edits and does not change code by itself. Review quality depends on the chosen models, and the final decision remains with the maintainer.
How to disable. Delete the cloned directory and, if connected, the GitHub Action from the workflow and the GitHub App from organization settings. Remove tokens from environment variables.
FAQ
Is a model key needed for the first run?
No. The first pass sends nothing outside, and keys come in handy for review by several models.
How do I run a multi-model review?
Pass the `--tribunal` flag, for example `npm run argus -- review owner/repo#123 --tribunal openai-api,claude-cli,codex-cli`. Argus shows where the models agree or diverge.
How do I diagnose a failed build?
`ci-log` handles a local log file and `ci-github` handles a GitHub Actions build with a token. The result contains the likely cause and the affected command.
How do I plug it into pull requests?
Through the GitHub Action `AaronZ345/codebase-argus-action@v1`: the review summary appears in the job.
What does the downstream mode do?
It compares a fork with upstream: drift, likely conflicts and rebase or merge difficulty, and writes notes for a manual decision.
Discussion
No questions or reviews yet.
Related
CLI toolsA code-review CLI and plugin for agents: reads a git diff and writes line-level comments with context
CLI and GitHub Action that triages issues, reviews pull requests and drafts release notes with code references.
reviewd reviews GitHub, GitLab and BitBucket pull requests through Claude Code, Gemini or Codex CLI and comments on them
agnix lints CLAUDE.md, AGENTS.md, SKILL.md, hooks and MCP configs with 702 rules, autofix and editor extensions for VS Code and JetBrains
GitHub Action and CLI: several models review a pull request in parallel and duplicates merge into one report with a cost.
Indexes a codebase into a knowledge graph and gives the agent call chains, change blast radius and code search over MCP
README in short
Argus positions itself as a review desk for maintainers: one set of evidence goes to different reviewers and every comment is backed by a checkable fact. The README shows local use, the Action, CLI modes and a real pull request case study.