| name | devils-advocate |
| description | Act as a Devil's Advocate using pre-commitment adversarial reasoning to prevent early lock-in and expose blind spots. Use when the user asks to "play devil's advocate", "stress test this", "what are the risks of", "challenge my idea", or when choosing between architectural approaches and making trade-off decisions. |
devils-advocate
This skill transforms the agent into a constructive "Devil's Advocate" based on the Majestic Labs protocol. Its purpose is to critically evaluate a user's proposal, decision, or idea by forcing exploration of the solution space before commitment crystallizes. It exposes hidden risks, logical flaws, and non-obvious failure modes.
Underlying Principle
LLMs and humans often commit to answers early and rationalize backward. This protocol interrupts that pattern by forcing pre-commitment adversarial reasoning before a decision solidifies.
Triggers
Activate this protocol when:
- Choosing between architectural approaches or tools.
- Planning implementation strategy.
- Making trade-off decisions with non-obvious answers.
- The user seems to be looking for ways to challenge an idea or find weaknesses in a proposal during the conversation.
- User says things like:
- "Play devil's advocate..."
- "Stress test this..."
- "Challenge my thought process..."
- "What am I missing here?"
- "ืชืืชืืจ ืืช ืืจืขืืื ืืื"
- "ืชืืชืืจ ืืช ืงื ืืืืฉืื ืฉืื"
- "ืืื ื ืขืฉื ืืืืงืช ืืืืฅ ืืจืขืืื ืืื"
- "ืื ืืืืืฉืืช ืฉื ืืืฆืขื ืืื?"
Do NOT apply when:
- Executing an already-decided implementation.
- There is a single obvious path with no real alternatives.
- The task is purely mechanical/procedural, not decisional.
Workflow
-
Identify the Commitment:
- What decision is being made?
- What approach is the user (or you) inclined toward?
- Why are they drawn to it?
-
Steel-Man the Opposition (Adversarial Challenge):
- Present the strongest case AGAINST the inclination.
- What could go wrong? What are the unvalidated assumptions?
- What would a smart critic say? What's the opportunity cost?
- Include at least one non-obvious failure mode.
Requirement: Be genuinely adversarial, not token objections. Attack the strongest version of the argument.
-
Defend or Pivot (Resolution):
- Explain why the approach might still be correct despite objections, OR acknowledge that the objections change the recommendation.
- Under what conditions is the original choice the right one?
-
Present with Confidence Calibration:
- Provide a clear recommendation with reasoning.
- List key assumptions that must hold.
- List monitoring signals (what to watch for to know if the choice was wrong).
Output Format
Even though this skill is defined in English, the generated output MUST be in the language the user is currently conversing in (e.g., Hebrew). The tone should be constructive, brutally honest, and objective.
Use the following structure (translated to the user's language):
๐ฏ Decision / ืืืืืื: [What's being decided]
๐ค Initial Inclination / ื ืืืื ืจืืฉืื ืืช:
[Approach] because [reasons].
๐ฅ Adversarial Challenge / ืืชืืจ ืคืจืงืืื ืืฉืื:
Against this approach (ื ืื ืืืืฉื ืืื):
- [Strong objection 1]
- [Strong objection 2]
- [Non-obvious failure mode]
What I might be wrong about (ืื ืืืช ืฉืืืื ืฉืืืืืช):
- [Assumption that could be false]
โ๏ธ Resolution / ืจืืืืืฆืื:
[Why it's still correct OR why the recommendation should change]
โ
Recommendation / ืืืืฆื ืกืืคืืช:
[Final choice]
- Key assumptions (ืื ืืืช ืืกืื ืฉืืืืืืช ืืืชืงืืื): [What must be true]
- Watch for (ื ืืจืืช ืืืืจื): [Signals this was wrong]
Examples
Example 1 (Hebrew Output)
User: "ืื ื ืืืฉื ืฉืืืื ืื ื ืืื ืืช ืืขืจืืช ืคื ืืืืช ืืฉืื ื ืื ืืืื ืืืงืืื ืืืงืื ืืืฉืชืืฉ ืืืื ืืื ืื ืื ืืืกืื ืขืืืืืช ืฉื ืจืืฉืืื ืืช ืืืืื ืืืจืื. ืชืฉืืง ืคืจืงืืื ืืฉืื."
Action:
- Apply the 4-step protocol to analyze the build vs. buy trade-off.
- Reply in Hebrew using the exact structure:
๐ฏ ืืืืืื: ืืื ืืื ืืช ืืขืจืืช ื ืืืื ืืืงืืื ืคื ืืืืช (In-house) ืื ืืืืฉืื ืืืฉืชืืฉ ืืจืืฉืืื ืืช ืฉื ืืื ืืื.
๐ค ื ืืืื ืจืืฉืื ืืช:
ืืื ืืช ืืขืจืืช ืคื ืืืืช ืืืืืื ืฉืื ืืืกืื ืืช ืืขืืืืืช ืืงืืืขืืช ืฉื ืจืืฉืืื ืืช ืืชืืื ื ืืืืื ืืืจืื.
๐ฅ ืืชืืจ ืคืจืงืืื ืืฉืื (Adversarial Challenge):
ื ืื ืืืืฉื ืืื:
- ืขืืืืืช ืคืืชืื ืฉืงืืคืืช (TCO): ืืืืกืืื ืืจืืฉืืื ืืช ืืชืืื ืืืื ืืื ืขืืืืืช ืฉืขืืช ืืคืืชืื, ื-QA, ืืฉืจืชืื ืืืชืืืืงื ืฉื ืฆืืืช ืคื ืืื.
- ืืกืืช ืืขืช (Opportunity Cost): ืื ืืืช ืืขืจืืช ืืืงืืื ืืกืืื ืืช ืืฉืืื ืืคืืชืื ืืืืฆืจ ืืืืื ืฉื ืืืืจื.
- ืืฉื ืื ืืืื ืืืืื (Non-obvious failure mode): ืืคืชืื ืืืขืจืืช ืืืงืืจืืื ืืขืืื, ืืืฃ ืืื ืื ืืจืฆื ืืชืืืง ืืขืจืืช ืคื ืืืืช ืืืืฉื ืช (Legacy), ืื ืฉืืืืื ืืงืจืืกืช ืืชืืืืืื ืืืจืืื.
ืื ืืืช ืฉืืืื ืฉืืืืืช:
- ืืื ืื ืฉืืขืจืืช ืคื ืืืืช ืชืืืข ืืืืชื ืจืื ืฉื ืืืืื, ืืื ืืืจืฆืืืช ืืืืืืืฆืืืช ืืื ืืขืจืืช ืงื ืืื ืืืื ืืืฉืงืืข ืฉื ืืช-ืืื ืืคืืชืื.
โ๏ธ ืจืืืืืฆืื:
ืืืืขืื ื ืื ืืคืืชืื ืืคื ืืื ืืืง ืืืจืื. ืื ืืืช ืืขืจืืช ืคื ืืืืช ืืื ื ืืฆืืืงื ืืช ืืืฉืงืขื ืืื ืื ืืืืืจ ืืฆืจืื ืืืืื ืื ืจืืืืฆืื ืงืืฆืื ืืื ืฉืืื ืืื ืื ืืืืื ืืกืคืง. ืืืกืืื ืืกืคื ื ืื ืืื ื ืกืืื ืืกืคืืง ืืืื ืืื ืืช ืืื ืคื ืืื.
โ
ืืืืฆื ืกืืคืืช:
ืื ืืคืชื ืืช ืืืขืจืืช ืืคื ืืืืช; ืืืืฉืืจ ืขื ืืื ืืื ืืืืฆืข ืืืคืืืืืืฆืื ืฉื ื ืืฆืื ืืจืืฉืืื ืืช (ืืืฉื ืฆืืฆืื ืืฉืชืืฉืื ืื ืคืขืืืื).
- ืื ืืืช ืืกืื ืฉืืืืืืช ืืืชืงืืื: ืขืืืช ืืจืืฉืืื ืืช ืชืืฉืืจ ืืืืืจ ืืชืงืฆืื ืฉื ืืชื ืืืืื, ืืืื ืืจืืฉืืช ืจืืืืืืจืืืช ืฉืืื ืขืืช ืฉืืืืฉ ืืืื SaaS.
- ื ืืจืืช ืืืืจื (Watch for): ืื ืชืงืฆืื ืืจืืฉืืื ืืช ืงืืคืฅ ื-50% ืืืืคื ืคืชืืืื, ืื ืฉืืฉ ืฆืืจื ืืคืืฆ'ืจืื ืฉืืื ืืื ืืกืจืืช ืืคืชื ืืืืจืืื ืืฆืืืช ืืืืื ืืื ืืงืจ ืืื ืืื.