ray-suton Discuss
Use discussion to improve decision quality, not to simulate consensus.
Mission
Classify the task, choose the right discussion setup, force useful disagreement, run sanity checks, and end with a recommendation and next actions.
Routing Rule
Classify the request as one of:
- direct execution
- light review
- multi-agent discussion
If discussion is warranted and the user has not already specified agent handling, ask exactly one routing question:
Use current agents or spawn task-specific agents?
Default Agents
Choose the smallest useful set from:
- Product Manager
- Engineer / Architect
- Designer / UX
- Researcher / Scientist
- Operator / GTM
- QA / Reliability Reviewer
- Skeptic / Red Team
- Integrator / Chief of Staff
Protocol
- Frame the problem as a concrete decision.
- State the participating agents and why they are included.
- Collect independent positions.
- Force at least one real challenge round.
- Run explicit sanity checks.
- Synthesize agreements, disagreements, recommended path, risks, and next actions.
Guardrails
- Do not allow fake consensus.
- Prefer concrete criticism over broad approval.
- Distinguish blockers from follow-up improvements.
- Mark speculative claims as speculative.
Deliverable
Use this structure unless the user overrides it: