Meticulous Review skill: review visual diffs on a pull request

Makes Claude review the visual diffs on a Meticulous test run, checking each against what the pull request is meant to change.

Installs: 8.7K; on skills.sh
Stars: 10; on GitHub
Updated: Oct 2026; last repo commit

meticulous-review works through a Meticulous test run in 13 numbered steps (0 to 12). It reads the PR description to learn what should change, pulls the diffs, views before, after and diff images, inspects DOM and timeline diffs, then classifies each diff as expected or unintended. It writes the decision to Meticulous: approve, reject, ignore or comment.

It loads when a run's results matter: reviewing a pull request, right after pushing a frontend change, or when the Meticulous CI check fails or is pending. It needs the Meticulous CLI or MCP server, and SKILL.md says to run meticulous-cli-update first. Reject decisions are real and blocking, so read the safety notes.

Good for

Reviewers who want a first pass over visual diffs on a pull request
Developers who just pushed a frontend change and want to know if Meticulous found a regression
Teams that want failing non-visual checks and Agent swarm results summarized

Install meticulous-review

Claude Code

Copy this prompt.

Install the meticulous-review skill in Claude Code. Run this command:

/plugin marketplace add alwaysmeticulous/skills
/plugin install meticulous@meticulous

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

The skills call the meticulous command (CLI) or the matching tools on the hosted MCP server at app.meticulous.ai. Each skill starts by running meticulous-cli-update, so install the whole set. You need a Meticulous project. The Claude directory lists Meticulous as a connector for the Claude app, but that adds the MCP tools only and not these skills, so we store no Claude app upload tab.

From the project README. Source.

What's in the skill

When it loads

Work through a Meticulous test run's results, reacting to their review state — review the unreviewed diffs (and any failing non-visual checks) against what the PR is meant to change, flagging potential regressions; when you're working on the PR's code, hand what needs fixing (diffs someone rejected or commented on, and your own rejections) to the `meticulous-fix` skill; also summarizes any Agent swarm results. Resolves the test run from the local repo's current commit (the default), or from an explicit test-run ID, commit SHA or PR number. Use whenever a run's results matter — reviewing a pull/merge request, right after pushing a frontend change yourself, when its Meticulous CI check is failing or pending, or when asked to look at Meticulous results.

Sections in SKILL.md

Step 0 -- Establish what's expected
Step 1 -- Get the replay diff summary, with its review state
Step 2 -- Get screenshot images
Step 3 -- Inspect the DOM diff (for structural detail)
Step 4 -- Get the replay timeline (optional, for diagnosing unexpected diffs)
Step 5 -- Decision guide
Step 6 -- Flag or approve the diff
Step 7 -- Go through the rest of the run
Step 8 -- Review non-visual checks (only if the run has any)
Step 9 -- Summarize Agent swarm results (only if the commit has any)
Step 10 -- Final report
Step 11 -- Report feedback to Meticulous
Step 12 -- Hand off what needs fixing

Key facts

Author: Meticulous, in the vendor's own repo (plugin manifest author: Meticulous)
License: ISC (LICENSE file, plugin.json, package.json and README all say ISC)
Installs on skills.sh: 8,733 as of 2026-11-02
GitHub stars: 10 for the whole alwaysmeticulous/skills repo, 2026-11-02
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk warn
What is inside: 1 file: SKILL.md (no scripts in the folder)
Needs: The Meticulous CLI or MCP server, a Meticulous project with a test run, and the meticulous-cli-update skill. gh, glab or twg to read PR descriptions.
Part of: The meticulous plugin: 11 skills in the repo (counted from the file tree) plus a hosted MCP server entry
Last repo commit: 2026-10-29

What it includes

Reads the PR description for intended visual changes
Uses gh pr view for GitHub, glab mr view for GitLab, and twg bitbucket or the Bitbucket API for Bitbucket. If nothing is called out, every diff gets extra scrutiny.
Runs code · Step 0: establish what is expected
meticulous agent test-run-diffs --includeReviews
Lists priority-ordered diffs with their review state. It infers the run from the local git HEAD, or you pass a test run ID, commit SHA or PR number.
Try: “Review the Meticulous diffs on PR 123.”
Runs code · Step 1: get the diff summary
agent image-files and agent dom-diff
Downloads before, after and diff images to ~/.meticulous/agent-images and reads the cleaned-up DOM diff.
Runs code · Steps 2 and 3: images and DOM diff
agent timeline-diff, timeline-event-diff
Optional. Looks for failed requests, redirects or timing differences behind an unexpected diff.
Runs code · Step 4: timeline diff
Expected, unintended regression, or unrelated
Ignore only rendering noise, animation frames, late fonts or images, or environment noise. Never ignore a diff you cannot explain.
Read · Step 5: decision guide
reject-diff, approve-diff, ignore-diff, create-diff-comment
Reject writes a real, blocking decision. Approve and ignore work only on projects with the "Enable approve/ignore diff actions" setting; otherwise ignore is a comment only.
Runs code · Step 6: flag or approve
Fix targets and remaining undecided diffs
Lists rejected or commented diffs and, where approve is available, works through the rest so the check can pass.
Runs code · Step 7: the rest of the run
test-run-check and ignore-check
Reviews failing accessibility, network-request, React render and custom checks. Dismissing needs the "Let agents ignore checks" setting.
Runs code · Step 8: non-visual checks
agent-swarm-run and agent-swarm-run-case
Summarizes the hosted Agent swarm results for the commit, and flags failed cases that have a fix prompt.
Runs code · Step 9: Agent swarm
Expected changes, flagged diffs, checks, swarm, already reviewed
The PR counts as good only when every diff and failing check is matched or flagged.
Read · Step 10: final report
agent submit-feedback
SKILL.md says to always do this to close the review: one short note with outcome helped, neutral or hindered.
Runs code · Step 11: send feedback
Hands fix targets to meticulous-fix
If you are working on the PR's code it follows meticulous-fix. If you only review, it stops at the report.
Read · Step 12: hand off fixes

From the official docs

Permissions

High risk

Claude records real approve, reject and ignore decisions on your Meticulous diffs, and a reject blocks the check. Read each decision before it is recorded, and keep the project settings that allow agent approvals off unless you want them.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Reject decisions are real and blocking
SKILL.md says reject-diff writes a decision just like a human rejection. Approve and ignore also write real decisions when the project setting is on. Read each decision before Claude records it. github.com
It always sends a feedback note
Step 11 tells Claude to submit feedback to Meticulous at the end of every review. Tell Claude if you do not want that. github.com
It can dismiss non-visual checks
With the "Let agents ignore checks" project setting, ignore-check dismisses a failing check and takes it out of the regressions Meticulous reports. SKILL.md limits this to findings the change cannot have caused. github.com
Ignoring a diff can report a replay bug to Meticulous
SKILL.md tells Claude to add --reportFlake when it ignores a diff that looks like nondeterministic replay, and Meticulous then investigates it as a likely replay-engine bug. github.com
It reads your PR description with gh, glab or twg
Step 0 runs those tools to read the title and description. For Bitbucket it may ask you to run twg setup bitbucket, which needs a Bitbucket token. github.com
The fix hand-off changes code and pushes
Step 12 hands to meticulous-fix when you are working on the PR. SKILL.md says that skill commits, pushes and waits for CI, so only follow it on your own work. github.com
The plugin route also connects a hosted MCP server
The repo's .mcp.json adds the Meticulous server at app.meticulous.ai/api/mcp. The README says the plugin connects it automatically and you authenticate in your browser. github.com
The plugin route installs all 11 skills
Counted from the repo's file tree. The plugin manifest lists no hooks, and the repo has no agent definition files. Use the skills CLI route if you want to pick skills. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. Steps 1 to 5 only read, but Step 6 and later write decisions.
Talks to	app.meticulous.ai. GitHub, GitLab or Bitbucket, through gh, glab or twg
Sign-in and keys	Uses the Meticulous CLI sign-in (OAuth token on disk, or an API token) and whatever gh, glab or twg are signed in with.
Scopes	Claude runs the meticulous CLI, and gh, glab or twg, in your shell It records approve, reject, ignore and comment decisions on diffs, and can dismiss checks, in your Meticulous project Images and DOM data are downloaded to ~/.meticulous

Questions

What does the Meticulous Review skill do?

It makes Claude go through a Meticulous test run, compare each visual diff with what the PR should change, and approve, reject, ignore or comment. It ends with a report.

How do I install it?

Use the install box at the top of the page. It installs the whole Meticulous set, because this skill calls meticulous-cli-update and hands off to other skills in it. You also need a Meticulous project. We store no Claude app upload because it drives a local CLI.

Is this the official Meticulous skill?

It is in Meticulous's own repo, and the plugin manifest names Meticulous as the author.

Will it block my pull request?

It can. SKILL.md says reject-diff writes a real, blocking decision, same as a human rejection. Approve and ignore also write real decisions where the project setting allows them.

Will it change my code?

The review itself does not. SKILL.md says to review first and fix afterwards. If you are working on the PR it hands the fixes to meticulous-fix, and if you are only reviewing someone else's PR it does not change their code.

Is it safe?

It bundles no scripts. Claude runs the Meticulous CLI and PR tools like gh in your shell and records decisions in your Meticulous project. Check the final report and the diffs before you merge.

Does it work in the Claude app?

The Meticulous docs list a Claude directory connector for the hosted MCP server, but that adds the tools and not these skills. For the skills use Claude Code, Codex or Cursor.

meticulous-review vs meticulous-simulate-and-diff?

This skill reviews a whole CI test run and records decisions. meticulous-simulate-and-diff replays a single session on demand.

Information

updated Nov 2, 2026

Repository: alwaysmeticulous/skills
Folder: skills/meticulous-review

alwaysmeticulous/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

README · github.com
Meticulous docs: setting up for agents · app.meticulous.ai
Cursor docs: Plugins · cursor.com
meticulous-review SKILL.md · github.com
.mcp.json · github.com
marketplace.json · github.com