reviewlisted
Install: claude install-skill yusufkaracaburun/ai-kit
# Code Review
Review the diff against the base branch before merge. Inspired by [gstack /ai:review](https://github.com/garrytan/gstack), stripped of gstack-specific tooling.
Read `CONTEXT.md` and relevant ADRs first; they're cached truth. See [`context-discipline.mini.md`](../../../standards/rules/context-discipline.mini.md).
## Run mode
- **Claude Code (preferred):** delegate the full review pass to the `reviewer` subagent via the Task tool with `subagent_type=reviewer`. Pass: base branch, target ref (default `HEAD`), mode (`daily` or `comprehensive`), security depth (`default` or `deep`). Return the subagent's markdown report to the user verbatim, then discuss the verdict.
- **Cursor / hosts without subagents:** run the inline checklist below in the main context. The checklist below is the canonical source of truth — `reviewer`'s system prompt mirrors it.
For wide cross-file impact analysis where you don't need a full verdict, prefer `explore` (read-only sweep) over inline grepping.
## Process
1. **Scope** — identify changed files (`git diff --name-only` vs base branch).
2. **Understand** — read CONTEXT.md and relevant ADRs if they exist.
3. **Review** for:
- Correctness and edge cases
- Security (injection, auth, secrets, unsafe defaults)
- Test coverage for behaviour changes
- Naming aligned with CONTEXT.md domain language
- No drive-by refactors mixed with the stated change
4. **Report** — blockers first, then suggestions. Be specific (file + line).