How to use the Plan Review agent skill: install and run it

What is an agent skill, and what does Plan Review make the agent do?

An agent skill is a folder containing a SKILL.md file. The file tells an AI agent how to handle a specific kind of task. When the skill is loaded and a request matches it, the agent follows the instructions in that file.

Plan Review turns the agent into a multi-perspective plan reviewer. It reads the feature material, then evaluates it through three lenses so gaps, scope drift and missed opportunities surface before a feature moves into development.

• Strategy review: whether the plan fits company growth goals, which personas it serves, how it compares to competitors, whether the scope is right, and what the team gives up by building it.
• Design and UX review: six dimensions scored 0-10 (information architecture, interaction states, user journey, consistency, accessibility, and AI integration if the feature uses AI).
• Engineering feasibility: whether acceptance criteria are specific enough that a developer will not need to guess, plus data model implications, dependencies, performance and scale, and whether the scope fits the implied timeline.
• A structured report with blocking, important and suggested issues, recommended actions, and praise for what is working.

Which files does the review read?

The skill tells the agent to read every relevant document in the feature folder before reviewing. The main document is SPEC.md. Other files are used when they exist.

• validation.md: office hours verdict, demand evidence and wedge scope (if present).
• background.md: customer drivers and competitive context (if present).
• SPEC.md: the main document under review.
• supporting/: mockups, data files and visual references.

How do you install the Plan Review skill?

Download the plan-review ZIP from this page. The package contains one file, SKILL.md. There are two ways to add it.

Option 1 is for Claude Code users. Unzip the folder into your personal skills directory, ~/.claude/skills/, or into a project's .claude/skills/ folder if you want it available only in that project.

Option 2 is for the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal install: unzip into ~/.claude/skills/ so the skill is available across your projects.
• Project install: unzip into .claude/skills/ inside the project repository so teammates working in that repo get it too.
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP.

How do you trigger the skill with a natural request?

You do not need a special command. Describe what you want in plain language, and point the agent at the feature folder that contains your plan. The skill's description covers reviewing a feature plan or requirements doc before development begins, so requests that match that purpose should load it.

Name the document if you have more than one. Mention the feature folder if your SPEC.md sits alongside validation.md or background.md, because the skill reads those too.

• Review the SPEC.md in features/bulk-export before we hand it to engineering.
• Run a plan review on this requirements doc and list the blocking issues first.
• Check my feature brief for missing UX states and vague acceptance criteria.
• Critique this plan from a strategy, design and engineering angle and tell me what to fix before dev starts.

What does the report look like?

The agent writes a report titled Plan Review for the feature. It has a score for strategy and for engineering, a design table with a score and notes for each UX dimension, and an issues list sorted into three levels. Blocking items must be fixed before development. Important items should be fixed but do not stop work. Suggestions are optional.

The report ends with recommended actions. The skill also points to a companion exec-summary skill, which can produce an executive summary saved as executive-summary.md in the feature folder once the issues are resolved.

• Blocking: must fix before dev, each tied to a specific requirement.
• Important: should fix, not blocking.
• Suggestions: nice to have.
• Recommended actions: specific steps before development.

What are the safety notes and requirements?

The skill's listed requirements are none. It does not run scripts and does not use the network, according to the package details. It is a set of instructions in a single SKILL.md file.

Because the review is written by an AI agent, its scores and findings are opinions. Treat them as a structured checklist, not a sign-off. The skill also does not know about company-specific context it has not been given, such as real personas or competitor data, so supply those files if you want the review to reference them.

• No listed requirements.
• Runs scripts: no. Uses network: no.
• Licence: Apache-2.0.
• Give the agent background.md and competitor material if you want checks against real context.