Review code changes across alignment, correctness, security, quality, testing, docs, and style areas.
npx skills add https://github.com/alvis/.agents --skill review-code-alvis
Please help me install this Agent Skill. Skill: review-code Source: https://github.com/alvis/.agents/tree/main/plugins/coding/skills/review-code Command: npx skills add https://github.com/alvis/.agents --skill review-code-alvis
open|fixed|acknowledged|deferred|skipped statuses, persisted under the active work root in review.md and reviews/*.md, surviving partial reruns. - Pattern confirmation and GitHub handoff: Outstanding findings are grouped into patterns, confirmed with the user inline or via interactive discovery, and verified against the remote default branch before suggesting GitHub issue creation. - Use Case: After implementing an authentication refactor, run the review against the current diff to verify alignment with the approved plan, catch security and correctness defects, and confirm each problem pattern before handing eligible ones off as GitHub issues. ## Quick Start Ask the assistant to review the current changes with /review-code --work-id=<your-work-id> and confirm the reported problem patterns.High-intent search queries and answers about installing and using this skill.
Run the review with a specifier such as a file path, directory, glob, PR number, or git range plus a work ID. The reviewer compares changed behavior against the canonical state.md task definitions and reports drift, omissions, and defects per area.
Pass the --area parameter with a comma-separated list such as security,correctness or testing. Area selection controls coverage only; one holistic reviewer still authors all selected area reports.
No. The review is strictly read-only: reviewers never edit reviewed code, run builds or linters themselves, or write outside their assigned report files. Remediation belongs to a separate fix workflow.
Yes, but only for user-confirmed problem patterns verified as still present on the remote default branch at an exact commit SHA. Issue drafting and publication are handed off to a dedicated issue skill after explicit user authorization.
Findings use stable IDs and dispositions that persist across reruns. A finding closes only with verified fix evidence or a justified risk disposition, and reopened findings must cite new evidence invalidating the prior disposition.
Completed findings keep their status but the affected area is treated as stale if a reviewed task's immutable definition changed. The area must be re-reviewed before the review can close.
ArthTech-AIDispatch fresh subagents for independent implementation tasks and review spec compliance before code quality.
open-mercatoCreates follow-up GitHub issues from PRs with spec detection and labeling rules.
seal-harnessReviews Haskell code for correctness, idiomatic style, performance pitfalls, and best practices.
SriRamkunamsettyAutomate pre-commit code verification with static security scans and linting.