Skip to content

Review a repository with focused evidence

Anatomize helps agents and people inspect an unfamiliar repository, localise a change, review duplication, and verify the result. It produces compact Markdown for review and stable, versioned JSON for automation.

python -m pip install anatomize
anatomize review start . --format json --output review/session.json
anatomize review dossier review/session.json \
  --profile orientation --format markdown --output review/orientation.md

Run the complete quickstart Browse common workflows

Choose what you need to do

Review a repository

Capture the current checkout, find its entry points, and create a focused dossier for a file, symbol, test, document section, workflow, or diagnostic.

Start the review

Integrate an agent

Use versioned JSON through the CLI, the high-level Python API, or the optional read-only MCP server. Expand partial results without repeating the review.

Integrate automation

Import results from another tool

Add saved SARIF, JUnit, coverage, mutation, duplication, workflow, dependency, or research results. Authors of new integrations can use the advanced provider format.

Import tool results

Complete a review

  1. Run review start once for the checkout.
  2. Use review dossier with the profile and target that match the task.
  3. Inspect the status, conflicts, omissions, and limitations before editing.
  4. Use review similarity and review consolidate when assessing duplicate code, tests, or documentation.
  5. Record implementation obligations with review intent.
  6. Capture the changed checkout and run review change and review verify.
  7. Attach the focused Markdown dossier and closure report to the code review.

The lifecycle guide provides the commands for each step. The artifact gallery explains which outputs are intended for people and which are intended for machines.

Read a result correctly

  • complete means the required evidence was available within the requested scope.
  • partial identifies useful evidence and the material omissions that remain.
  • blocked means the current session cannot answer the question defensibly.
  • A similarity candidate is something to review, not an instruction to merge.
  • Closure requires current evidence for every declared obligation.

Use interpret review results for the complete reading order and troubleshooting for stale, incompatible, partial, or unavailable evidence.

Keep the review boundary explicit

Anatomize does not execute repository code or approve changes. Run the repository's own tests, linters, documentation checks, security tools, and research validations, then import their saved results where they help the review. Source text is excluded from sessions unless explicitly requested.

For exact commands, schemas, and public Python names, use the public reference.