Review code changes with specialized reviewer roles and confidence-scored findings.
npx skills add https://github.com/ronaknnathani/relay --skill review-ronaknnathani
Please help me install this Agent Skill. Skill: review Source: https://github.com/ronaknnathani/relay/tree/main/skills/review Command: npx skills add https://github.com/ronaknnathani/relay --skill review-ronaknnathani
--comment. - Use Case: Before opening a pull request, run the review to catch a swallowed exception, an untested error path, and a guard removed that git history shows was added deliberately — each with a concrete fix. ## Quick Start Ask the AI to review the current branch's changes against the base branch and produce a severity-ranked findings report before opening the pull request.High-intent search queries and answers about installing and using this skill.
Run the review skill, which delegates the diff to focused roles like code-reviewer, silent-failure-hunter, security, and pr-test-analyzer. Each role scores findings for confidence and only items scoring 80 or higher survive into the final severity-ranked report.
Invoke the review with the --comment flag to post a mandatory summary comment plus one inline comment per surviving finding via the gh CLI. Line numbers must fall inside the PR diff range or the GitHub API returns a 422 error.
Yes, the default report mode works entirely locally using git diff against a pinned merge-base, producing a severity-ranked findings report without posting anything. The gh CLI is only needed for PR-scoped diffs, comment deduplication, and inline mode.
The should-review gate skips when the diff is empty, the PR is closed, the change is trivially correct like a dependency bump, or the exact HEAD commit was already reviewed with no new commits. The skip reason is always reported.
Every finding is scored 0-100 for confidence verified against the actual code, and anything below 80 is dropped. Pre-existing issues on untouched lines, linter-enforced style, and intentionally suppressed lines are also excluded before reporting.
Avoid it for trivially correct changes like single config values or dependency bumps, which the gate skips anyway. It is advisory and never blocks merges, so it does not replace human judgment for architectural decisions.
AO-HySReview branch diffs against repository standards and specs with ranked actionable findings.
hypeJunctionIterate on open pull requests to resolve CI failures and review feedback.
NealschIdentifies assets, trust boundaries, threats, and mitigations during system architecture design.
jhannkaRuns parallel blind dual-judge adversarial code reviews with fix and re-judgment cycles.