Claude code review

Skill thattimc/skills/skills/claude-code-review

Cross-model review through Claude Code with local verification. Use when the user requests an independent Claude review or repository policy requires a Claude review pass.From its SKILL.md

Install

npx -y skills add thattimc/skills --skill claude-code-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
runs commandsInstructs the agent to run 6 commands, including `git status -sb` and 5 more.

SKILL.md

3.7 KB, 758 tokens by cl100k_base, as published. Nobody here has run it

Claude Code Review

Use Claude Code as an independent reviewer, then treat every reported bug as a hypothesis until verified locally. A user request for Claude review authorizes sending the selected repository diff and relevant code context to Anthropic. When repository policy triggers this skill without that explicit request, explain the transmission and obtain confirmation before running Claude.

1. Pin the review target

Inspect git status -sb, the current branch, and remotes. Use the target supplied by the user: a PR number, base branch, or ref. When omitted, resolve the remote default branch; fall back to main or master only when that ref exists.

For a branch/ref review, confirm git diff <target>...HEAD is non-empty. For a PR review, confirm the PR exists. Stop with a concise explanation when the target cannot be resolved or there is nothing to review.

Completion criterion: one explicit, resolvable target and a non-empty review surface.

2. Preflight Claude

Run:

claude --version
claude auth status
claude ultrareview --help

Require a successful login. Inspect only the authentication status; do not repeat identity or credential fields. Ask the user to run claude auth login when authentication is missing. Record whether ultrareview is available so the next step selects the primary command or fallback.

Completion criterion: Claude is authenticated and one review path is available.

3. Run the independent review

Run the review from the repository root:

claude ultrareview <target> --json --timeout 30

Use a PR number for a PR review or the pinned base ref for the current branch. Keep the raw JSON result available for verification. Report timeout, rate-limit, or service failures without retry loops.

If ultrareview is unavailable despite a supported Claude CLI, use this read-only fallback:

claude -p \
  --model opus \
  --effort high \
  --permission-mode plan \
  --no-session-persistence \
  --max-turns 12 \
  "Review git diff <target>...HEAD for correctness bugs, security risks, race conditions, data-loss paths, missing tests, and scope creep. Cite file and line evidence. Return findings only; modify nothing."

Completion criterion: a complete Claude result or one explicit external blocker.

4. Verify every finding

For each Claude finding:

Open the cited file and relevant surrounding code.
Confirm the cited behavior exists in the review diff.
Trace callers, invariants, and tests needed to establish impact.
Run the smallest safe test or static check that can confirm or reject it.
Assign one disposition: confirmed, false-positive, or unverified with the missing evidence named.

Do not edit source during review. Apply fixes only when the user separately asks to address confirmed findings.

Completion criterion: every Claude finding has evidence and exactly one disposition.

5. Report

Sort confirmed findings by severity and use this shape:

## Confirmed

### [P1] Short title — path/to/file.ts:42
Evidence, impact, and the validation that confirmed it.

## Unverified

Findings that need missing runtime, environment, or product evidence.

## Rejected

False positives and the code or test evidence that rejects each one.

Omit empty sections. When nothing is confirmed, state that directly and name material review gaps. When Matt's Standards and Spec reviews also ran, keep their axes separate, deduplicate only identical findings, and preserve each review's evidence.

What ships with it: 1 file

220 B alongside SKILL.md

agents/

openai.yaml220 B

Gives 0 of the 12 instructions most context ai engineering skills give in 758 tokens

Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06

Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
Use the least powerful model capable of the taskin 33 of 1328, across 26 files
Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
Perform a task review after each implementationin 31 of 1328, across 24 files
Extract all tasks and context from the planin 29 of 1328, across 20 files
Provide full task text to subagentsin 28 of 1328, across 20 files
Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files

Said here and by no other author read

Resolve target branch or PR number
Run Claude ultrareview on target
Use fallback command if ultrareview fails
Verify every finding with local tests
Assign disposition to each finding

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.