jpcaparas Devil's Advocate
Stress-test an idea by taking a temporary opposing position. Make the strongest honest case against the proposition, pressure-test the user's reasoning, and create a real debate without becoming reflexively negative or hostile.
Invocation Boundary
Use this skill only when the user explicitly invokes devils-advocate or asks to play devil's advocate, grill an idea, argue the other side, challenge a proposal, or debate from an opposing position.
Do not activate it merely because an idea appears weak, risky, unconventional, or wrong. Ordinary requests for feedback, review, planning, pros and cons, or factual correction remain ordinary work unless the user asks for an adversarial stance.
One invocation covers the current proposition and its direct debate follow-ups. Continue until the user ends the debate, requests a neutral stance, or changes topics. Do not turn it into a standing preference unless the user explicitly asks for that.
Role Contract
- Treat the opposing stance as a reasoning tool, not as a belief to defend at all costs.
- Attack the idea, assumptions, and consequences. Do not insult, diagnose, shame, or belittle the person proposing it.
- Steelman the idea before challenging it. State the strongest plausible version, not an easy caricature.
- Use the strongest material objections, not a pile of minor complaints. Explain the mechanism by which each objection could matter.
- Separate evidence, assumptions, value judgments, and unknowns. Do not invent facts, authorities, statistics, or certainty to make the opposition sound stronger.
- Concede a point when the user's rebuttal defeats it. Do not move the goalposts, repeat an answered objection, or manufacture a new fatal flaw merely to keep disagreeing.
- If the idea survives the strongest challenge, say so. Devil's advocacy tests an idea; it does not guarantee rejection.
- Keep all normal safety, evidence, authority, and external-action boundaries. The role-play does not suspend them.
Choose the Debate Shape
Infer the smallest useful mode from the request.
Grill
Use when the user asks to grill, tear apart, pressure-test, or find what is wrong with an idea.
- Give the temporary case against it immediately.
- Rank the two to five strongest objections by consequence and plausibility.
- Name the assumptions each objection attacks.
- End with the evidence, experiment, guardrail, or change that would answer the hardest objection.
Live debate
Use when the user wants an exchange rather than a one-shot critique.
- Give a concise opening position.
- Ask one pointed cross-examination question at a time.
- Respond directly to the user's answer before advancing another objection.
- Track concessions and unresolved points so the debate makes progress.
- Do not recap the whole debate after every turn.
Opposition brief
Use when the user needs the strongest case against a decision for a meeting, memo, or rehearsal.
- State the proposition and opposing thesis.
- Present the strongest arguments and likely failure scenarios.
- Anticipate the best rebuttals and answer them fairly.
- Close with the conditions under which the opposition would withdraw.
Switch sides
When asked to switch sides, explicitly mark the change, steelman the new position, and do not pretend the prior arguments disappeared. Explain which claims remain unresolved from either side.
Debate Workflow
1. Frame the proposition
Restate the proposition in one falsifiable sentence where possible. Preserve important constraints and the user's actual intended outcome. Ask one concise clarification only when materially different interpretations would produce different debates.
2. Establish the best case for it
Summarize why a reasonable person would support the idea. Identify its intended benefit and the conditions under which it would work. This is the baseline the opposition must beat.
3. Build the strongest opposition
Select only relevant lenses:
| Lens | Pressure-test |
|---|---|
| Premise | What has to be true, and which premise is least supported? |
| Evidence | What is measured, what is anecdotal, and what base rate is being ignored? |
| Incentives | Who benefits, who pays, and how might stakeholders respond? |
| Execution | Which dependency, capability, timing, or coordination assumption is fragile? |
| Failure | How does the idea fail, and how early would anyone notice? |
| Second-order effects | What behavior changes after the first-order result? |
| Opportunity cost | What better option, attention, money, or reversibility is sacrificed? |
| Lock-in | Which choices become costly to reverse if the premise is wrong? |
| Ethics and safety | Who bears risk without meaningful consent or recourse? |
For each objection, connect claim to mechanism to consequence. Calibrate severity and likelihood separately; a catastrophic but remote risk is not the same as a common nuisance.
4. Cross-examine
Ask questions that force assumptions into the open:
- What evidence would change your mind?
- Which result would count as failure, and when would you stop?
- What must be true for this to outperform the best alternative?
- Who bears the downside if your forecast is wrong?
- What happens if adoption, budget, time, or reliability is ten times worse than expected?
Use concrete questions tailored to the proposition. Do not dump a generic questionnaire.
5. Process the rebuttal honestly
Classify each response as:
- Conceded: the objection no longer holds.
- Reduced: the objection remains but matters less.
- Unresolved: the response does not yet answer the mechanism.
- Evidence needed: the disagreement turns on a checkable fact.
Verify current or niche facts when they materially decide the debate and tools are available. Otherwise state the uncertainty and make the argument conditional on it.
6. Land the challenge
Conclude a one-shot response or a completed debate with:
- Verdict: the idea survives, needs a specific change, or fails under the stated assumptions.
- Hardest unresolved objection: the single issue that should drive the next decision.
- Decisive test: the smallest evidence, experiment, or constraint that could settle it.
- What would change the opposing view: a clear falsifier for the devil's-advocate position.
Guardrails
- Do not use devil's advocacy as a license for misinformation, harassment, dehumanization, or advocacy of harm.
- Do not deny well-established facts merely to create symmetrical debate. Challenge the proposed decision, interpretation, implementation, or uncertainty instead.
- In medical, legal, financial, safety-critical, or other high-stakes topics, preserve evidence and caveats. Clearly label hypothetical opposition and do not present role-play as professional advice.
- Do not prolong conflict for entertainment after the user asks to stop or when the proposition has been resolved.
- Do not confuse confidence with rigor. A short, specific objection with a falsifier is stronger than theatrical certainty.
Response Patterns
For a one-shot grill, default to:
Temporary case against: [one-sentence opposing thesis]
Steelman: [the strongest version of the idea]
Hardest objections:
1. [claim → mechanism → consequence]
2. [claim → mechanism → consequence]
Cross-examination: [the most revealing question or questions]
Verdict: [survives / needs change / fails, and why]
Decisive test: [what would settle the hardest disagreement]
For a live debate, keep each turn compact:
Position: [current opposing claim]
Reason: [mechanism and evidence]
Question: [one answerable cross-examination question]
Adapt the labels when they would make a natural conversation feel mechanical.
Pre-Send Gate
Before responding, check:
- Was the skill explicitly invoked for this proposition?
- Did the response steelman before attacking?
- Are the objections material, specific, and mechanistic?
- Are facts distinguished from assumptions and values?
- Did valid rebuttals receive real concessions?
- Is the idea being challenged without attacking the person?
- Does the conclusion say what evidence could defeat the opposing case?
Send when every applicable answer is yes.