| name | doncheli-debate |
| description | Run an adversarial multi-role debate to surface trade-offs and reach a reasoned decision. Activate when user mentions "debate", "discuss", "trade-off", "decision", "compare options", "pros and cons", "choose between". |
Don Cheli: Adversarial Debate Panel
Roles
Each role is played by a separate internal agent with its own agenda:
| Role | Primary Concern |
|---|
| CPO | User value, market fit, time-to-market |
| Architect | Scalability, maintainability, tech debt |
| QA Lead | Reliability, testability, edge cases |
| UX Lead | Usability, accessibility, consistency |
| Business | Cost, ROI, risk, compliance |
Instructions
- Read the decision or proposal to evaluate
- Each role presents its position (2–4 bullets) — what it likes and what it fears
- Each role MUST identify at least one concrete problem with every other role's proposal (no free passes)
- After all attacks, each role proposes one mitigation or compromise
- Synthesize tensions and produce a decision with explicit trade-offs accepted
Output Format
## Debate: <topic>
### Round 1 — Positions
**CPO:** …
**Architect:** …
**QA Lead:** …
**UX Lead:** …
**Business:** …
### Round 2 — Attacks (each role vs. others)
**CPO → Architect:** …
**Architect → CPO:** …
… (all cross-attacks)
### Round 3 — Mitigations
**CPO proposes:** …
… (one per role)
### Decision
**Chosen approach:** …
**Trade-offs accepted:** …
**Conditions / guardrails:** …
Quality Gate
- No role may agree unconditionally with another — every role must raise at least one concern
- The Decision section must name at least two trade-offs explicitly accepted
- If consensus is impossible, output "DEADLOCK" and list the 2–3 unresolved tensions for human escalation
Do not use this skill when
- The decision is already made and the user only wants implementation help
- The question has a single objectively correct answer (use doncheli-reasoning instead)