reviewd CLI
reviewd reviews GitHub, GitLab and BitBucket pull requests through Claude Code, Gemini or Codex CLI and comments on them
- 70 ★ on the repository
- Last commit
Broad access
Why this level
- Write-capable PR tokens give the tool the right to post comments and approve PRs on your behalf
- The AI CLI runs with access to the local clone and executes the commands from test_commands
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/reviewdDetects 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.
- Write-capable PR tokens give the tool the right to post comments and approve PRs on your behalf
- The AI CLI runs with access to the local clone and executes the commands from test_commands
This is third-party code. Review the repository files before installing.
What it does
reviewd is a local assistant for pull request review. It takes open PRs from GitHub, GitLab and BitBucket, creates a git worktree for each in your own clone and runs an installed Claude Code, Gemini or Codex CLI with a structured prompt and a JSON schema. Then reviewd posts inline comments and a summary comment with severity levels. Before the AI call it runs your linters, type checks and tests from `test_commands`, and failures go into the analysis. New commits on a PR trigger a fresh review and old comments are cleaned up. State lives in SQLite, so commits already reviewed are skipped. There are two modes: `reviewd pr my-project 42` reviews one PR and `reviewd watch` runs a loop across all repositories in an open terminal. Filters skip drafts, small diffs and PRs with patterns in the title, and optional auto-approve fires on thresholds for size, severity and finding count. The `reviewd init` wizard guides first setup.
Who it is for. Teams on GitHub, GitLab or BitBucket who hold a Claude Code, Gemini CLI or Codex subscription and keep local clones of their repositories.
Good fit when
- Colleagues send many PRs and you want an AI first pass before your own review, with no CI and no cloud service
- Claude reviews some repositories and Gemini others, with the model set per repository
- Before review you need to run the linter and tests from `test_commands` and show failures to the AI
- A critical finding should create a BitBucket task and block the merge
- You want to check a fresh PR without posting: `reviewd pr` with `--dry-run` shows the result in the terminal
Not a fit when
- You need review without a local clone: reviewd works only with your clones
- Review should run in the cloud on every commit automatically: the `reviewd watch` loop lives in an open terminal
- The machine has no Python 3.12 and none of the Claude, Gemini or Codex CLIs
- You require a manual security checklist: the AI gives findings but does not promise completeness
Example request
Review PR 42 in my-project with reviewd and --dry-run and show findings by severityLimitations
- Python 3.12 or newer and one of the CLIs (`claude`, `gemini`, `codex`) with authentication are required.
- Each host needs a token: a GitHub fine-grained token with Pull requests read and write, a GitLab token with the api scope, or a BitBucket token.
- Claude CLI on a headless server loses its interactive login, so long-running use needs a token from `claude setup-token` (valid one year) or `ANTHROPIC_API_KEY`.
- `reviewd watch` runs only while the terminal stays open.
How to disable. Stop the `reviewd watch` loop with Ctrl+C. The package is removed with `pip uninstall reviewd` or `uv tool uninstall reviewd`, and the config sits in `~/.config/reviewd/config.yaml` and `.reviewd.yaml` in repositories, removed by hand.
FAQ
How do I install reviewd?
Run `pip install reviewd` or `uv tool install reviewd`, then `reviewd init`. The wizard finds clones, detects the host, validates tokens and writes the config.
Which models and CLIs does reviewd support?
Claude Code, Gemini CLI and Codex CLI. The `cli` key sets the choice globally or per repository, and a model can be set there too.
Does reviewd post comments right away?
Yes, in normal mode it posts inline and summary comments on the PR. Add `--dry-run` to preview the result first.
What does reviewd do with drafts and small PRs?
In `watch` mode drafts are skipped unless the title contains `[review]`, `[claudiu]`, `[ask]` or `[bot review]`. Diff size and title patterns are configurable too, and the `pr` command reviews a PR regardless of draft status.
Can reviewd approve PRs itself?
Yes, if auto-approve is on: it fires on thresholds for diff size, severity and finding count and on AI-evaluated rules. The approval rationale lands in the summary comment.
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
Two skills: a pull request review where two models argue, and a helper that works through review comments round by round
CLI with a virtual board of AI experts that reviews a PR or local diff and returns a score, severity level and evidence.
A Claude Code plugin: Claude does the task, then a stop hook starts an independent Codex review and Claude fixes the findings in the same session
The CLI matches linter diagnostics to code changes and reports findings locally or through an approved review channel.
Three skills work through Greptile review comments on a pull request, fix them and re-run the review until none are left open
README in short
The page lists capabilities: local clones, worktrees, parallel reviews, running tests, structured output and smart re-reviews. Install, the setup wizard for three hosts, the pipeline and global and per-project config follow.