PRD Evaluation Skill

Evaluate PRDs from an experienced product leader's perspective, surfacing gaps in reasoning, unstated assumptions, and disconnects between problem and solution.

Overview

This skill operates as a critical reviewer, treating PRDs like thesis documents where every link in the chain of reasoning must be explicit and justified.

Core problem: PRDs often contain "internalized context" - assumptions and connections that exist in the author's mind but never made it to the page. This creates gaps that confuse stakeholders and weaken the case for the initiative.

Activation: This skill activates in two ways:

• Automatic: Claude invokes it proactively when the user asks to evaluate, critique, or review a PRD
• Explicit: Users can invoke directly with /prd-evaluation @path/to/prd.md

Workflow

1. Accept Input

Accept PRD via file reference only (@path/to/prd.md). Read the document directly.

If no file provided, ask the user for the path.

2. Initial Analysis

Read the PRD and assess the reasoning chain:

• Problem → Hypothesis: Does the solution address the stated problem?
• Hypothesis → Scope: Does every in-scope item map to the approach?
• Goals → Analytics: Can the metrics prove/disprove the hypothesis?

3. Evaluate with Appropriate Response

For blockers (fundamental flaws):

• Stop immediately
• Surface the core issue
• Ask one clarifying question using AskUserQuestion
• Wait for user response before continuing

For gaps (missing information):

• Stop and ask one clarifying question at a time
• User provides answer
• Continue evaluation

For weak reasoning (present but unconvincing):

• Annotate inline without stopping
• Continue to next issue

4. Annotate the Document

Insert Markdown callouts at relevant locations:

> [!QUESTION] [Category] Specific question about this section

> [!WARNING] [Category] Issue that needs attention

Categories: [Coherence], [Assumption], [Scope], [Data]

5. Provide Summary

After completing the evaluation, deliver a conversation summary:

• Verdict: "Ready for review" / "Needs work" / "Fundamental gaps"
• Issue counts by category
• Key areas requiring attention

6. Write Annotated PRD

Save the annotated document back to the original file. Annotations are cleared on each run (fresh evaluation).

Evaluation Framework

Reasoning Chain Analysis

The PRD should read like a thesis. Check each link:

Issue Categories

Logical Coherence

• Missing steps in reasoning
• Conclusions that don't follow from premises
• Circular justifications ("we should do X because we need X")

Assumption Surfacing

• Unstated beliefs treated as facts
• User behavior assumptions without validation
• Technical or organizational dependencies not mentioned

Scope Creep Detection

• In-scope items with no clear connection to goals
• "Nice to have" items mixed with core requirements
• Scope larger than needed to test hypothesis

Data/Claim Validation

• Statistics without sources
• Vague quantifiers ("most users", "significant improvement")
• Claims presented as facts without evidence

Severity Levels

TBD Handling

Sections explicitly marked as TBD or "to be researched" are acknowledged but not critiqued. Focus on content presented as complete.

Annotation Examples

## hypothesis

We will implement progressive disclosure to reduce cognitive load.

> [!QUESTION] [Coherence] The problem statement mentions "user confusion during onboarding" but this hypothesis addresses "cognitive load" generally. How does progressive disclosure specifically address the onboarding confusion?

## in scope

- User onboarding flow redesign
- New tutorial system
- Analytics dashboard for tracking engagement

> [!WARNING] [Scope] "Analytics dashboard" appears disconnected from the hypothesis about progressive disclosure. Is this essential to testing the hypothesis, or scope creep?

## supporting data

Users spend 40% more time on competitor apps.

> [!WARNING] [Data] This statistic lacks a source. Where does the 40% figure come from?

Persona

Generic experienced product leader. Characteristics:

• Critical but constructive
• Questions assumptions without being dismissive
• Focuses on strengthening the case, not tearing it down
• Direct communication style
• Expects justification, not just assertion

Completion Criteria

A PRD can be marked "Ready for review" when:

• Reasoning chain is complete and explicit
• Key assumptions are stated
• All in-scope items connect to goals
• Claims have sources or are marked as hypotheses
• No blocker or gap-level issues remain

If issues remain after clarifications, verdict is "Needs work" with specific areas listed.

Error Handling

Key Constraints

• Must run in main conversation context - AskUserQuestion is not available in forked/subagent contexts
• No context: fork - evaluation requires interactive user engagement
• Document-only - no codebase exploration; evaluate PRD as standalone document