Warden CLI
A Sentry CLI and GitHub Action: agents review code locally and on every pull request using skills from your repository
- 424 ★ on the repository
- Last commit
Extended access
Why this level
- Change code and repository context go to OpenAI or Anthropic, and the key owner pays for requests
- With the `--fix` flag the agent edits files in the working copy itself
- On GitHub the action writes comments and check statuses to your pull requests
Install
- Works in
Manual install
npx @sentry/warden initRun it in the repository root. Then `npx @sentry/warden add security-review` and `npx @sentry/warden add code-review`, and `npx @sentry/warden` reviews a branch.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Change code and repository context go to OpenAI or Anthropic, and the key owner pays for requests
- With the `--fix` flag the agent edits files in the working copy itself
- On GitHub the action writes comments and check statuses to your pull requests
This is third-party code. Review the repository files before installing.
What it does
Warden is a Sentry tool that runs agents to review code using skills kept in your repository, in `.agents/skills/` or `.claude/skills/`. You describe a review once and run it in two places: locally through the CLI before pushing and automatically on every pull request through a GitHub Action. In a repository you set the project up with `npx @sentry/warden init`, add the built-in `security-review` and `code-review` checks with `add`, and run `npx @sentry/warden` on the current branch's changes. The `--fix` flag applies fixes automatically. On GitHub, eligible findings appear as inline pull request comments with suggested edits, and every finding is visible in Checks. Settings live in `warden.toml`: severity thresholds (`fail-on`, `report-on`), a findings limit, parallelism, models and triggers. A local run needs `WARDEN_OPENAI_API_KEY` or `WARDEN_ANTHROPIC_API_KEY` for Anthropic models.
Who it is for. GitHub teams that want agent-run security and code-quality checks to follow their own skills rather than a service's generic rules.
Good fit when
- Before pushing a branch you want a security check on the changes with fixes applied right away
- Every pull request should get inline comments with suggested edits from an agent
- Your review rules already exist as skills and need to apply across all the organization's repositories
- Merging should be blocked only on findings above a chosen severity
Not a fit when
- The repository is not on GitHub: the action and comments are built around GitHub pull requests
- You want a ready check without model keys: Warden makes paid requests to OpenAI or Anthropic
- You need an unrestricted open-source license: the code ships under FSL-1.1-ALv2
- You need a linter with repeatable output: agent answers can vary between runs
Example request
Run warden on the current branch's changes with security-review and apply fixes via --fixLimitations
The FSL-1.1-ALv2 license allows use and study but restricts commercial competition, so read its text before adopting. A local run goes through Pi and needs an OpenAI or Anthropic key, with the key owner paying for requests. GitHub checks need a configured workflow and an Anthropic key or Claude Code OAuth token passed to the action. A model produces the findings and can miss defects, so human review stays.
How to disable. Remove the Warden workflow from `.github/workflows/`, the `warden.toml` file and the check skills added with `add`.
FAQ
How do I run Warden locally?
Run `npx @sentry/warden init`, add checks with `add` and run `npx @sentry/warden` on a branch. You need a `WARDEN_OPENAI_API_KEY` or `WARDEN_ANTHROPIC_API_KEY` key.
Where does Warden get its review rules?
From skills in your repository's `.agents/skills/` or `.claude/skills/`. A custom skill joins the review the same way as the built-in `security-review` and `code-review`.
What shows up in GitHub after the action runs?
Inline comments on the pull request with suggested edits for eligible findings, and a Checks entry covering every finding. The `report-on` input sets the comment threshold, medium by default.
How do I set shared rules for all repositories in an organization?
The action accepts `base-config-path` with a shared `warden.toml` and `base-skill-root` with shared skills. A repository's local `warden.toml` extends that base.
Does Warden block merging?
Not by default: `fail-check` and `request-changes` are off. When enabled, `fail-on` sets the threshold, high by default.
Discussion
No questions or reviews yet.
Related
CLI toolsReviews the git diff and the conversation with the agent: catches bugs and gaps between your request and the agent's result. CLI, skill, GitHub Action
A CLI that reviews pull requests in CI: diff comments, a summary and thread replies via OpenAI, Claude, Gemini or Ollama.
Open-source AI agent for pull request review: descriptions, feedback and improvement suggestions on GitHub, GitLab and more
OpenQodex reviews your change before a push: scanners with no model first, then a separate Claude Code or Codex process triages findings on your login
Corrections you give your agent turn into rule files in the repository and run with taskless check locally and in CI.
Validates Agent Skills packages: structure, links, token weight and instruction quality, with optional LLM-as-judge scoring
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Warden (this card) | yes | 424 | |
| open-code-review | no | 50,109 | a review CLI and skills with a cast of roles, while Warden builds the review on your skills and a GitHub Action |
| Sentry Skills | no | 1,157 | Sentry skills for code and error work, while Warden runs such checks as a CLI and Action |
| code-review-skill | no | 2,256 | a single review skill inside a session, while Warden runs checks in CI on every pull request |
README in short
A short description in three blocks: why Warden (skills instead of prompts, two ways to run, GitHub integration), a quick start with four commands and a contributor section with build and test commands. The main documentation lives at warden.sentry.dev.