Pre-push Review

Use the strongest locally available reviewer to catch problems while they are still cheap to fix. Run this before the first push and again before every later push to an existing PR.

This is the local counterpart to douwebot: a high-judgment standards review paid for by the developer's model subscription. It is independent of the automatic CI Review, which runs on GitHub after CI passes.

Read the review rubric

Read the prompt: of the douwebot review job in .github/workflows/bots.yml. It is the source of truth for what to look for, how to prioritise concerns, and what makes feedback useful. Apply its review judgment and comment-quality rules, but ignore its hosted workflow mechanics: triggers, checkout, model selection, pre-gathered file paths, and GitHub comment tools.

Read the root AGENTS.md, agent_docs/index.md and its relevant topic guides, plus every directory-specific AGENTS.md governing a changed file.

Gather local and PR context

First run gh pr view for the current branch.

• If a PR exists, read its title, body, base branch, linked issue, comments and reviews. Review the entire branch diff against that base, not just the latest commit. Use the existing discussion to avoid duplicate findings and to detect concerns that remain unresolved after an iteration.
• If no PR exists, use main as the base and review against the task context available locally. Skip only PR metadata that does not exist; scope and readiness are still valid review concerns.

Gather the corresponding local state:

git status --short
git merge-base <base> HEAD
git diff <base>...HEAD --stat
git diff <base>...HEAD -W
git diff HEAD

The last command includes staged and unstaged work that has not reached HEAD. Read a large diff in chunks, core implementation before tests, and skip generated files (uv.lock, cassettes).

Return the review locally

Do not post comments, submit a GitHub review, or modify the branch. Return only actionable findings as text: file:line, the problem, and the concrete fix. Put higher-level concerns before lower-level ones, following the ordering in the douwebot rubric. Say plainly when there are no findings.