Devil's Advocate

Parallel advocate/critic analysis that surfaces tensions and decision factors for a given topic.

Topic: $ARGUMENTS

1. Gather Context

Search the codebase for code, documentation, and configuration related to the topic (Glob/Grep to locate, Read up to 5 key files). Identify:

• Current state and constraints
• Stakeholders and affected systems
• Prior art or existing patterns in the codebase

2. Run Parallel Sub-agents

Use the Agent tool to launch both agents simultaneously in a single message (two Agent tool calls in parallel). Each agent receives the gathered context but cannot see the other's output.

For each agent, read the prompt template from ${CLAUDE_SKILL_DIR}/agents/ and substitute {{topic}} and {{context}} with actual values.

3. Synthesize

After both agents return, combine their outputs into the following format. Do not editorialize — present both sides faithfully, then identify where they conflict.

Output Format

## Advocate

[Advocate agent output — arguments in favor]

## Critic

[Critic agent output — arguments against, risks, alternatives]

## Key Tensions

| Advocate claims | Critic claims |
| --------------- | ------------- |
| [Position A]    | [Counter A]   |

## Decision Factors

[What information or validation would resolve each tension — actionable next steps]

Common Mistakes

• Topic too broad (e.g., "frontend strategy" → narrow to "adopting React Server Components for the dashboard")
• Editorializing in synthesis — present both agents' outputs faithfully before identifying tensions
• Ignoring Decision Factors — the actionable next steps are the most valuable part, not just the debate itself