How do I install the Review skill?

Run npx skillmds@latest add sujeet-pro/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 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 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 Review free to use?

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

Who published Review?

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

Review by sujeet-pro — Agent Skill

Name: Review
Author: sujeet-pro

Review

Review, audit, sanity-check, look-at any review-able target. Triggers on a GitHub PR URL (fetched via the gh CLI), a local path or "." (review the working tree), a markdown/doc file, or a comment-thread URL. Read-only by default; --fix applies accepted findings locally and pushes to the PR branch after confirmation (never force, never merge, never to a protected branch). Produces severity-tiered findings (blocker / critical / should / may / nit) with path:line and <=15-word verbatim evidence quotes. Six dimensions: correctness, tests, security, performance, readability, consistency. For a deep PR review with cross-file code-context retrieval, use /adk:pr-review.

sujeet-pro c50ce08 5 files · 14.2 KB Updated Sep 17, 2026

File contents

review — review any target (lightweight)

Polymorphic on the target. Read-only by default. Lightweight by design — no worktree, no embeddings. For a deep PR review with code-context retrieval and inline comment posting, use /adk:pr-review.

The full operating contract lives in this skill folder — read these as you need them:

Aspect	File
How you review (voice, tiers, evidence rules)	`persona.md`
The phased process + Workflow orchestration	`workflow.md`
Hard rules + refusals + safety	`rules.md`
Target routing (PR / local / doc / thread)	`dispatch.md`

Quick start

Read dispatch.md and classify the target. Resolve the diff:
- GitHub PR URL → gh pr view <url> --json … for metadata, gh pr diff <url> for the diff. (GitHub access is always the gh CLI — assume the user ran gh auth login.)
- . or local path → git diff against the merge-base of the current branch.
- doc / comment thread → fetch the content directly.
Read persona.md — adopt the findings-first reviewer stance.
Run the workflow in workflow.md. For any non-trivial diff, fan the six dimensions out in parallel with the Workflow tool (one code-reviewer / security-auditor agent per dimension), then adversarially verify each finding before it survives. Small diffs (≤ ~150 LOC, single concern) may be reviewed inline without a workflow — but still one dimension at a time.
Report severity-tiered findings with path:line + ≤15-word quotes, then a one-line ship | iterate | reject recommendation. Posting to a PR or applying --fix is gated per rules.md.

Workflow is the default for real diffs

"Always have a workflow." A diff worth reviewing gets the parallel-dimension Workflow in workflow.md: review fans out → each finding is adversarially verified by an independent skeptic → only survivors are reported. This catches what a single linear pass misses and kills plausible-but-wrong findings. Skip the Workflow only for a trivial diff, and say so.

Modes

default — review, report findings, recommend. Nothing is posted or changed.
-i — walk each finding with the user (accept / reject / edit) before it lands in the report or gets posted.
--fix — after the review, apply accepted findings locally, validate, and push to the PR's head branch after explicit confirmation. Never force-push, never merge, never push to a protected branch (rules.md).
--deep — use a stronger reasoning profile; auto-select for security-sensitive, cross-module, or ambiguous diffs.

---name: reviewdescription: Review, audit, sanity-check, look-at any review-able target. Triggers on a GitHub PR URL (fetched via the gh CLI), a local path or "." (review the working tree), a markdown/doc file, or a comment-thread URL. Read-only by default; --fix applies accepted findings locally and pushes to the PR branch after confirmation (never force, never merge, never to a protected branch). Produces severity-tiered findings (blocker / critical / should / may / nit) with path:line and <=15-word verbatim evidence quotes. Six dimensions: correctness, tests, security, performance, readability, consistency. For a deep PR review with cross-file code-context retrieval, use /adk:pr-review.---# review — review any target (lightweight)Polymorphic on the target. **Read-only by default.** Lightweight by design — no worktree, no embeddings. For a deep PR review with code-context retrieval and inline comment posting, use `/adk:pr-review`.The full operating contract lives in this skill folder — read these as you need them:| Aspect | File ||---|---|| How you review (voice, tiers, evidence rules) | `persona.md` || The phased process + Workflow orchestration | `workflow.md` || Hard rules + refusals + safety | `rules.md` || Target routing (PR / local / doc / thread) | `dispatch.md` |## Quick start1. **Read `dispatch.md`** and classify the target. Resolve the diff: - **GitHub PR URL** → `gh pr view <url> --json …` for metadata, `gh pr diff <url>` for the diff. (GitHub access is always the `gh` CLI — assume the user ran `gh auth login`.) - **`.` or local path** → `git diff` against the merge-base of the current branch. - **doc / comment thread** → fetch the content directly.2. **Read `persona.md`** — adopt the findings-first reviewer stance.3. **Run the workflow in `workflow.md`.** For any non-trivial diff, fan the six dimensions out in parallel with the **Workflow tool** (one `code-reviewer` / `security-auditor` agent per dimension), then adversarially verify each finding before it survives. Small diffs (≤ ~150 LOC, single concern) may be reviewed inline without a workflow — but still one dimension at a time.4. **Report** severity-tiered findings with `path:line` + ≤15-word quotes, then a one-line `ship | iterate | reject` recommendation. Posting to a PR or applying `--fix` is gated per `rules.md`.## Workflow is the default for real diffs"Always have a workflow." A diff worth reviewing gets the parallel-dimension Workflow in `workflow.md`: review fans out → each finding is adversarially verified by an independent skeptic → only survivors are reported. This catches what a single linear pass misses and kills plausible-but-wrong findings. Skip the Workflow only for a trivial diff, and say so.## Modes- **default** — review, report findings, recommend. Nothing is posted or changed.- **`-i`** — walk each finding with the user (accept / reject / edit) before it lands in the report or gets posted.- **`--fix`** — after the review, apply accepted findings locally, validate, and push to the PR's head branch *after explicit confirmation*. Never force-push, never merge, never push to a protected branch (`rules.md`).- **`--deep`** — use a stronger reasoning profile; auto-select for security-sensitive, cross-module, or ambiguous diffs.

Frequently asked questions