accountable-review
A Claude Code and Codex plugin that turns a pull request into a published HTML Review Map to help reviewers understand a change before merging.
At a Glance
accountable-review plugin/skills released under the MIT License. Running it requires your own Anthropic API key or Claude subscription, which is an external cost.
Engagement
Available On
Alternatives
Listed Nov 2026
About accountable-review
accountable-review is an open-source plugin from WyeWorks that turns a pull request into a Review Map: a published HTML page guiding a reviewer through what changed, how the new behaviour works, what existing code is affected, and what to understand before merge. It ships two skills, review-map and setup-ci, and installs into Claude Code and Codex. The repository's latest listed release is version 1.3.0.
What It Is
It is an AI-assisted code review tool aimed at what the README calls comprehension debt: code entering a codebase faster than a team can understand it. Rather than approving or scoring a PR, it helps a human reviewer build a mental model of the change. There is no approval verdict, severity score, or confidence badge; human judgment stays final.
The review-map skill supports Rails and Rust codebases (a Rust backend or Rust in general, with or without a separate client such as Next.js). Elixir/Phoenix and React support are described as still in development.
How a Review Map Is Built
The skill traces a PR around the behaviour being implemented rather than the order of files in the diff, including unchanged code that gains new meaning. A page contains a What changed paragraph, optional Context, What needs your attention (checkpoints framed as questions), a Read the code in this order list, Impact outside the diff chains, and a collapsed Evidence and diff coverage block.
Claims are labelled by how they are known (from unchanged code, inferred from tests, inferred, uncertain) and carry file:line citations, with collapsed source excerpts quoted from the repository. At the default --effort high, read-only claim-falsifier subagents try to contradict each analysis note before it reaches the page; --effort low skips that pass. --mentor adds framework primers for reviewers new to a stack, and --update re-reads only new commits and edits an existing map in place.
Running It and CI
Run it inside the repository against the current branch, a PR number, a PR URL, or a branch name. In Claude Code the result is a Claude Artifact; in Codex or other agents it is a local HTML file, and --output <dir> writes static HTML to a chosen directory. The setup-ci skill writes one GitHub Actions workflow that generates a map for review-ready pull requests, uploads it as an artifact, and posts a single comment linking it. It skips drafts, forks, bots, and trivial changes. CI execution currently uses Claude Code and Anthropic credentials; Codex CI is not supported yet.
Optional settings live in .accountable-review.yml, covering effort, mentor mode, context, theme, and CI options.
Limits to Know
The README states that a Review Map is a pass rather than an audit, has non-random blind spots (configuration, data, queues, other services), and can be wrong in places, so citations should be checked before acting. It depends on the model behind it, and only Claude Code and Codex are tested.
Community Discussions
Be the first to start a conversation about accountable-review
Share your experience with accountable-review, ask questions, or help others learn from your insights.
Pricing
Open Source (MIT)
accountable-review plugin/skills released under the MIT License. Running it requires your own Anthropic API key or Claude subscription, which is an external cost.
- MIT License
- review-map skill produces a Review Map HTML page from a pull request
- setup-ci skill configures CI to generate a map on each review-ready PR
- Supports Claude Code and Codex
- Requires user-supplied ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN (external cost)
Capabilities
Key Features
- Generates a Review Map HTML page from a pull request
- Traces changes around behaviours rather than file lists
- Surfaces impact on unchanged code with attached excerpts
- Labels claims by evidence source with file:line citations
- Independent claim-falsifier subagents at high effort
- Mentor mode with framework primers
- Update mode to edit a map in place as a branch moves
- setup-ci writes a GitHub Actions workflow that publishes maps as artifacts
- Optional .accountable-review.yml configuration
- Supports Rails and Rust codebases, with Phoenix and React in development
