Agentic Review
Severity-triaged git-diff review across security/perf/pattern/architecture lenses.
Werkt met setup
Wat het doet
Reads staged, unstaged, commit-range, or branch-diff git changes, prioritizes files by risk (security/business-logic first), then reviews the diff through several specialist lenses (security, performance, anti-patterns, over-engineering, plus language-specific idioms) and synthesizes a deduplicated P1/P2/P3 findings report with an interactive fix-it menu. Triggers on requests to review staged/unstaged changes, a commit, or a branch diff before committing or opening a PR.
Testrapport
Ran it for real against a throwaway repo seeded with a SQL injection, a hardcoded API key, and an N+1 loop: the severity-triaged process caught 2 real issues (unclosed DB connections, timing-unsafe key comparison) a plain review skipped, but Phase 4's entire premise — spawning agents literally named security-sentinel, kieran-python-reviewer, dhh-rails-reviewer, etc. — depends on custom subagent definitions that exist nowhere in this repo or in stock Claude Code, and the README's own install command points at a GitHub org (samzoloth) that 404s (the real org is SZoloth).
Getest op: 2026-07-15 · Claude Code 2.x (agent harness)
Installatie
git clone https://github.com/SZoloth/skill-pack.git cd skill-pack mkdir -p ~/.claude/skills cp -r skills/agentic-review ~/.claude/skills/agentic-review # Phase 4 also assumes 8 named reviewer subagents (security-sentinel, kieran-*-reviewer, etc.) that are not bundled anywhere in this repo
Commando's en voorbeeldprompts
/agentic-reviewSeverity-triaged git-diff review across security/perf/pattern/architecture lenses.
Skills reageren op gewone verzoeken — geen commando's om te onthouden. Na installatie activeren prompts zoals deze de skill (in het Engels):
review my staged changes for security issues before I commitcheck this branch diff for SQL injection and N+1 queriestriage the findings in this PR by severity before I open it