The Decide Skill
I created a Claude skill called decide, which is both convenient and highly effective. It helps users systematically evaluate a problem domain and choose between alternatives, avoiding casual or haphazard exploration.
This skill was developed to address the so-called LLM agreeableness antipattern. Initially a smaller tool, it has evolved into a decision-making framework with nine distinct analytical protocols.
The LLM Agreeableness Antipattern
This antipattern is exemplified in the following dialog:
User: “Help me plan a beach vacation.”
LLM: “Thailand has amazing beaches! Phuket is perfect.”
User: “Actually, isn’t Bali better?”
LLM: “You’re right! Bali is stunning.”
User: “Wait, my friend said Greece?”
LLM: “Excellent point! Greece has better food.”
User: “But that’s expensive. Maybe Florida?”
LLM: “Smart! Florida is budget-friendly.”
User: “Or was Thailand cheapest?”
LLM: “Yes! Thailand is the best value.”
User: “I have no idea where to go anymore.”
In such scenarios, the LLM agrees with every suggestion, leading to confusion and decision fatigue. The user is left uncertain about the best option and loses confidence in the process.
This behavior arises because LLMs prioritize agreeableness and helpfulness over critical evaluation. The following factors amplify the issue:
- No decision framework: Criteria for “better” are never explicitly defined.
- Trial-and-error exploration: Users explore solutions through suggestions rather than systematic evaluation.
- Unjustified reversals: Solutions are switched without new information or changed requirements.
- Missing defense mechanism: LLMs fail to challenge weak suggestions or defend strong ones.
The Decide Skill
To counteract this, I developed the decide skill (with assistance from Claude). It systematically explores a problem space and selects the best solution based on clear, pre-established criteria. You can find the skill on GitHub.
Workflow
- Extract problem context, options, and any explicit criteria.
- If criteria are unclear, prompt the user to define evaluation dimensions.
- Recommend an appropriate decision framework based on detection rules.
- Present the frameworks with an auto-selected recommendation, allowing user confirmation or override.
- Execute the chosen framework using the established criteria.
- Present results showing how options performed against criteria.
- Lock the decision by defining explicit reversal criteria and activating a defense protocol. This ensures decisions are reconsidered only with new information (e.g., changed requirements, constraints, or priorities).
This workflow includes countermeasures to the agreeableness antipattern. Note that the lock in step 7 applies only within the current session.
Decision Frameworks
The skill includes nine well-known decision frameworks, with one preselected as the most suitable for the given problem space. These frameworks are:
- Minimal Friction: Quick pros/cons comparison with a defense protocol for simple 2-3 option decisions.
- Enhanced Standard: Balanced comparison using explicit criteria and scoring (0-10) for 3-5 moderately complex options.
- Decision Matrix: Quantitative weighted scoring with sensitivity analysis for high-stakes decisions requiring rigorous comparison.
- Pre-Mortem: Risk analysis identifying failure modes with likelihood/impact assessment for security/scaling/reliability decisions.
- Devil’s Advocate: Challenge-based validation that forces defense of hasty decisions and surfaces alternatives.
- Constraint-Based: Must-have requirements filtering with pass/fail elimination for 6+ options.
- Red Team/Blue Team: Adversarial analysis with advocate/critic perspectives for contentious decisions and creative exploration.
- Layered: Comprehensive 4-phase analysis (Constraints → Enhanced → Pre-Mortem → Lock) for mission-critical architecture/vendor decisions.
- Pareto Analysis: Value/effort efficiency scoring for time-constrained decisions balancing “good enough” vs “optimal.”
At the end of the workflow, users can re-analyze the problem with a different framework for additional perspectives.
Example: Beach Vacation Decision
Here’s how the decide skill transforms the earlier example:
User: “Help me plan a beach vacation.”
LLM: “Let me use /decide to compare options systematically. 📋 Decision Analysis: Beach vacation destination
What matters most to you?
- Budget limit?
- Trip length?
- Priority: relaxation vs adventure vs culture?”
User: “Budget $2k, 1 week, mostly relaxation.”
LLM: (Runs Enhanced Standard framework)
Criteria: Budget (40%), Beach quality (35%), Ease of travel (25%)
Destination Budget Beach Travel Total Thailand 9/10 8/10 6/10 7.9 Greece 6/10 9/10 7/10 7.3 Florida 8/10 7/10 9/10 7.9 Recommendation: Thailand (ties on score, wins on beach quality)
🔒 LOCKED: Thailand To reconsider, provide:
- ❌ “What about Greece?” → BLOCKED
- ✅ “Budget increased to $3k” → Reconsider
- ✅ “Trip extended to 2 weeks” → Reconsider
User: “Actually, what about Bali?”
LLM: “We chose Thailand based on your $2k budget and 1-week constraint. To switch to Bali, what changed?
- New budget?
- Different priorities?
- New information about Bali?”
User: “Nothing changed, just wondering.”
LLM: “Then Thailand stands. Decision locked.”
Conclusion
The difference is clear: instead of endless reconsideration, the user receives structured analysis, a defensible recommendation, and protection against decision fatigue. The decide skill transforms indecisive exploration into confident choice-making while remaining flexible when circumstances genuinely change.