How to use the Review Implement Phase agent skill

What an agent skill is and what this one does

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a task matches the skill's description. This skill covers only the implementation phase of the review-framework workflow. It does not fetch review comments or decide which ones to address; earlier phases must produce that input.

For each action marked will_address, the skill directs the agent to make the code changes, run relevant checks, create focused commits, post an 'On it' comment when starting, post a 'Done' comment when finished, resolve the thread when applicable, and update the local status file.

• Reads review-actions.json and selects rows with decision will_address and status pending or in_progress
• Delegates the work to the review-implementer agent defined in agents/review-implementer.md
• Writes status back to review-actions.json and renders review-actions.md

Install the skill from the ZIP

Download the ZIP for this skill from this page and unzip it. To make it available for all your projects, place the folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.

You can also upload the skill in the Claude apps. Go to Settings, then Capabilities, then Skills, and upload the ZIP there.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your repository
• Claude apps: Settings → Capabilities → Skills

Check the requirements before you start

The skill needs Node.js and the GitHub CLI on your PATH. The GitHub account you use must be authenticated with permission to comment on pull requests and resolve review threads. The skill also needs network access to GitHub's API.

You must already have a valid review-actions.json file in the output directory, produced by the triage phase. If either node or gh is missing, the skill stops immediately and asks you to install it.

• node (Node.js) available on PATH
• gh (GitHub CLI) available on PATH and authenticated
• An existing review-actions.json from the triage phase
• Network access to GitHub's API

Trigger the skill with a natural request

You do not need to type a command. Ask the agent in plain language to implement the triaged review actions for a pull request, and include the PR URL. You can also name specific action IDs or files to limit the scope.

If you omit the output directory, the skill uses wip/reviews/<owner>_<repo>_pr-<number>/ by default. The skill's argument hint is [pr-url] [output-dir].

• "Implement the will_address items for https://github.com/owner/repo/pull/123"
• "Apply the triaged review fixes on this PR, only touch the files in src/api"
• "Carry out action R-004 from the review actions file for this PR"

Understand what the skill posts on GitHub

The skill posts real comments and resolves real threads on the pull request. It is designed to post standalone comments and thread replies only. It never creates a pending PR review, and it checks for a pending review authored by your account before starting and after each Done comment. If it finds one, it stops until that review is submitted or dismissed.

When the review item is a top-level review-body finding rather than an inline thread, the skill posts a top-level PR comment instead and does not try to resolve a thread.

• Thread replies are used for review_thread targets
• A top-level issue comment is used for pull_request_review targets
• status is set to done only after Done (and resolve, for review threads) succeeds

Review the output

When the run finishes, the skill reports the commits it created, the actions moved to done, and the artifacts it wrote. The artifacts are review-actions.json and review-actions.md in the output directory. Check the commits and the GitHub thread state before you consider the review closed.

• Commits created
• Actions transitioned to done
• review-actions.json and review-actions.md written