Spec Reviewer

Spec Reviewer

Parallel-agent review of spec documents. Produces a findings report without modifying the spec.

Workflow

1. Read the spec file provided by the user
2. Read AGENTS.md and architecture.md for project context and conventions
3. Create a team of 5 Explore agents — one per review dimension (see references/review-dimensions.md)
4. Wait for all agents to complete
5. Compile findings into a single report
6. Shut down the team

Team Setup

Create a team named spec-review. Spawn 5 agents using the Task tool with subagent_type: "Explore" and team_name: "spec-review". Run all 5 in parallel.

Each agent receives:

• The full spec content (inline in the prompt — do NOT tell agents to read files)
• The project guidelines from AGENTS.md and architecture.md (inline)
• Its specific review dimension and checklist from references/review-dimensions.md

Agent prompts must instruct: "Search the codebase thoroughly. Return a structured list of findings. Each finding must include: the spec section it relates to, the issue or suggestion, severity (critical/warning/info), and evidence (file paths, code snippets, or reasoning)."

Agent Assignments

Read references/review-dimensions.md for the full checklist to include in each agent's prompt.

Report Format

After all agents return, compile a single report using this structure:

# Spec Review: {spec name}

## Summary
{1-2 sentence overall assessment: is this spec ready, or does it need revision?}

## Critical Issues
{List any critical findings across all dimensions. If none, state "None found."}

## Findings by Dimension

### Codebase Consistency
{Bulleted findings from the consistency agent, each with severity tag}

### Code Reuse Opportunities
{Bulleted findings from the reuse agent}

### Performance
{Bulleted findings from the performance agent}

### Scope & Complexity
{Bulleted findings from the scope agent}

### Testability
{Bulleted findings from the testability agent}

## Verdict
{One of: READY / NEEDS REVISION / MAJOR REWORK, with a brief justification}

Omit dimensions with zero findings — don't include empty sections.