devil

Structured /devil command for stress-testing ideas against a devil's advocate.

Canonical names:

• Slash command: /devil <idea>
• Settings key: devil
• Extension dir: extensions/devil/

No agent-facing tool or shortcut is registered.

Quick Start

pi
/devil we should switch to microservices

Behavior

| Aspect | Behavior | |--------|----------| | Session state | Stateless. Each debate is independent. | | Conversation context | Automatically includes working dir, recent conversation, current contents of recently touched files. | | Model | Uses devil.model if configured, else current model. | | UI | Opens interactive progress UI, then result editor. | | Results | Edited verdict is appended as devil-result session state and submitted as user text. |

Configuration

~/.pi/agent/settings.json:

{
  "devil": {
    "model": "zai/glm-5.1",
    "maxRounds": 5,
    "stoppingConditions": "consensus",
    "concessionThreshold": 0.7,
    "onBlocking": "surface"
  }
}

CLI flags:

pi --devil-max-rounds 3 --devil-stop max_rounds

Output

Synthesis includes:

• recommendation: proceed, proceed_with_caution, revise, abandon
• blocking issues
• addressable concerns
• survived claims
• conceded weaknesses
• next steps
• debate metadata: rounds, blocking challenges raised, unresolved blocking issues in final verdict

Troubleshooting

"No model available"

Select model with /model, or configure devil.model.

"No API key"

Log in / configure key for selected or configured provider.

Irrelevant challenges

Make /devil <idea> more specific. Debate context is taken from current session and recently touched files.

Demo