| name | review |
| description | Run DeepWork Reviews on the current branch using .deepreview rules |
DeepWork Review
Run focused code reviews from .deepreview rules and DeepSchema-generated review rules. This native Pi package uses Pi commands and tools directly; it does not use MCP.
Routing
Stop and redirect when either condition applies:
- The user wants to configure, create, or modify review rules: use the
configure-reviews skill or /configure-reviews command when available.
- The user wants to create or update DeepSchemas: use the
deepschema skill.
Proceed only when the user wants to run reviews.
Running reviews can be expensive because it may generate and launch multiple focused reviewer agents. Run reviews only after a material set of changes has been completed, when the user explicitly asks for review, or when confirming a PR is in good form.
Preferred Command
Use /review when the command is available. The command generates native Pi review tasks, dispatches Pi subagents when available, and provides a sequential fallback when subagents are unavailable.
Manual Native Tool Flow
If this skill is loaded instead of the /review command:
- Call
deepwork_get_review_instructions.
- No
files argument reviews the current branch's detected changes.
- A
files list reviews only those files.
autostart_reviews_if_possible defaults to true; set it to false only when you need to inspect/validate generated review tasks without launching reviewers.
- If no rules are configured, ask whether the user wants to set up rules. If yes, use
configure-reviews.
- For each returned task, read the
prompt_file, follow it exactly, and inspect the listed files from the stated project root.
- Dispatch independent tasks to Pi subagents when available. Otherwise, run them one at a time in this session.
- Report findings with file paths and line references where possible.
- Apply obviously correct, low-risk fixes immediately.
- Ask the user before applying refactors, architectural changes, subjective style changes, or other risky fixes.
- Re-run applicable reviews after edits until no actionable findings remain or the user explicitly stops.
Changelog and PR Description Check
When relevant, run this alongside review tasks:
- Check for a changelog file.
- If a changelog exists and branch commits or changed files affect user-visible behavior, verify the unreleased/current section reflects the branch.
- If a PR is open and the changelog changed, verify the PR description matches.
Acting on Review Results
- Fix concrete correctness, typo, formatting, and unused-code findings when the fix is obvious.
- Ask before broad rewrites, public API changes, behavior changes, or debatable style changes.
- When a review passes, call
deepwork_mark_review_as_passed with the review ID from the task or prompt file.
- Do not mark a review as passed while actionable findings remain.
Iteration
After making a material set of fixes, call deepwork_get_review_instructions with files set to the files edited during the iteration when that is enough to re-check the relevant findings. If broader changes could affect other rules, run the review without a file filter. Avoid re-running after every tiny edit unless a review finding specifically requires immediate confirmation.