How do I install the Claude Review skill?

Run npx skillmds@latest add sandover/claude-review in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Claude Review safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Claude Review?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Claude Review free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Claude Review?

sandover (@sandover) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Claude Review by sandover — Agent Skill

Name: Claude Review
Author: sandover

Claude Review

Use the local Claude CLI as a second-opinion reviewer for briefs, plans, architecture, and code, then selectively incorporate its feedback. Use when the user invokes $claude-review or asks for an Opus or Claude review.

sandover ae5340d 2 files · 2.4 KB Updated Sep 17, 2026

File contents

Task Overview

You will seed a prompt to the claude CLI and read back the results; the claude CLI may use tools but its review should be read-only.

Default to claude-opus-5, or use claude-sonnet-5 when the user requests Sonnet. You remain responsible for deciding what feedback to incorporate

You should run claude from the same project directory you are working in so it sees the same AGENTS.md as you and the same directory context. Treat it as a peer agent. Claude may use tools.

Recipe

Choose target artifacts based on the user's request and context
Plan a brief prompt for claude. Claude is plenty smart; lots of specification doesn't help.
- Example: "Review the dirty working tree and propose improvements."
- Example: "Review ergo epic ABCDEF and propose improvements. Learn ergo via ergo --help"
Show the prompt to the user, for transparency.
Invoke claude, and then monitor it. You probably need to run claude outside the sandbox.
Evaluate claude's feedback carefully and objectively. Adopt the best of it.
Inform the user.

Example claude invocation command

claude -p \
  --model claude-opus-5 \
  --effort medium \
  --permission-mode auto \
  --tools Read,Glob,Grep,Bash \
  --no-session-persistence \
  --output-format json \
  "<review prompt>"

Use automatic permission mode instead of a command allowlist. It lets Claude run ordinary read-only shell discovery even when it adds wrappers, pipelines, or output limits, while retaining a permission boundary for risky operations. The prompt should keep the review read-only. Add other tools when the review needs them; do not use --allowedTools or bypass permission checks for a normal review.

Opus at medium effort can take several minutes. Poll at intervals no longer than 60 seconds and give the user concise progress updates. Do not interrupt a live process merely because it has not emitted partial output.

If claude supplies token, cache, or cost metadata, relay to the user, for transparency.

---name: claude-reviewdescription: Use the local Claude CLI as a second-opinion reviewer for briefs, plans, architecture, and code, then selectively incorporate its feedback. Use when the user invokes $claude-review or asks for an Opus or Claude review.---# Task OverviewYou will seed a prompt to the claude CLI and read back the results; the claude CLI may use tools but its review should be read-only.Default to `claude-opus-5`, or use `claude-sonnet-5` when the user requests Sonnet. You remain responsible for deciding what feedback to incorporateYou should run claude from the same project directory you are working in so it sees the same AGENTS.md as you and the same directory context. Treat it as a peer agent. Claude may use tools.## Recipe1. Choose target artifacts based on the user's request and context2. Plan a *brief* prompt for claude. Claude is plenty smart; lots of specification doesn't help. - Example: "Review the dirty working tree and propose improvements." - Example: "Review ergo epic ABCDEF and propose improvements. Learn ergo via `ergo --help`"3. Show the prompt to the user, for transparency.4. Invoke claude, and then monitor it. You probably need to run claude outside the sandbox.5. Evaluate claude's feedback carefully and objectively. Adopt the best of it.6. Inform the user.## Example claude invocation command```bashclaude -p \ --model claude-opus-5 \ --effort medium \ --permission-mode auto \ --tools Read,Glob,Grep,Bash \ --no-session-persistence \ --output-format json \ "<review prompt>"```Use automatic permission mode instead of a command allowlist. It lets Clauderun ordinary read-only shell discovery even when it adds wrappers, pipelines,or output limits, while retaining a permission boundary for risky operations.The prompt should keep the review read-only. Add other tools when the reviewneeds them; do not use `--allowedTools` or bypass permission checks for a normalreview.Opus at medium effort can take several minutes. Poll at intervals nolonger than 60 seconds and give the user concise progress updates. Do notinterrupt a live process merely because it has not emitted partial output.If claude supplies token, cache, or cost metadata, relay to the user, for transparency.

Frequently asked questions