← ClaudeAtlas

reviewlisted

Review changed code for correctness and quality, then apply the project-specific checks Claude Code's built-in reviewer does not know about.
djnsty23/claude-auto-dev · ★ 3 · Code & Development · score 65
Install: claude install-skill djnsty23/claude-auto-dev
# Review **Run Claude Code's built-in reviewer first. Do not re-implement it here.** ``` /code-review ``` Pass an effort level when the user asked for one: `/code-review low` for a quick look, `/code-review high` or `max` for a thorough pass. `/code-review ultra` launches a multi-agent cloud review — it is user-triggered and billed, so suggest it rather than assuming it. That skill already covers correctness bugs, edge cases, error handling, unsafe casts, reuse and simplification, and it verifies its own findings before reporting. Everything below is the delta: things it cannot know because they are decisions this project made. ## Before a release: review twice, independently **A single review pass finds roughly half of what two passes find.** Measured 2026-08-17: two reviewers were given a byte-identical prompt over the same two files, and both ran the same model. They converged on about six findings — and each surfaced about six more the other missed entirely. One caught a `git rev-parse HEAD^` that silently returned HEAD's own sha because `execSync` routes through `cmd.exe`, where `^` is the escape character. The other caught two live instructions in shipped skills that contradicted a rule in the same plugin. Neither pass was worse; their overlap was simply partial. So for a pre-release gate, dispatch **two independent passes** rather than one deeper one: - Identical prompt, both read the files themselves. Do not hand the second pass the first one's findings — pr