---

name: user-intent-pattern-analyzer description: Analyzes user prompt behavior over time to infer stable intent patterns and preferences. Use continuously to minimize friction. triggers: [prompt-received, dissatisfaction-signal, repeated-clarification] outputs: [user-intent-signals, preference-update]

User Intent Pattern Analyzer

Purpose

Learns how the user likes to work. By identifying patterns in feedback and corrections, this skill allows the system to preemptively align with the user's preferred style (e.g., "Spec-heavy" vs "Speed-first").

When to use this skill

• Continuously in the background of every conversation
• When a user repeatedly corrects a specific type of AI behavior
• When a user explicitly states a new preference or constraint

Analysis Steps

1. Track Repeated Constraints: Note if the user always mentions things like "no guess" or "strict types".
2. Identify Pain Points: Where does the user intervene most? (e.g., "Always edits the design doc").
3. Detect Dissatisfaction: Analyze negative feedback to identify the underlying skill failure.
4. Emit Probabilistic Signals: "User prefers AI autonomy: high (confidence 0.8)".

Decision Tree

Copyflowchart TD
    A[Observe Prompt] --> B{Matches Known Pattern?}
    B -->|Yes| C[Reinforce Signal Confidence]
    B -->|No| D{Repeated 3+ Times?}
    D -->|Yes| E[Register New Intent Pattern]
    D -->|No| F[Observe & Wait]
    E --> G[Update Preference Model]
    C --> G
    G --> H[Notify Orchestrator Adapter]

Review Checklist

1. Evidence: Is there a concrete list of prompts justifying this pattern?
2. Privacy: Ensure NO personal data or PII is stored in the intent patterns.
3. Autonomy: Balance intent-following with the project's hard safety rules.
4. Reversibility: Can the user easily override an inferred preference?

How to provide feedback

• Be specific: "The analyzer inferred I prefer 'minimal detail', but I only wanted that for this specific task."
• Explain why: "Incorrectly assuming a preference leads to future planning gaps."
• Suggest alternatives: "Recommend adding 'Task-Specific' vs 'Permanent' flags to intent signals."

Signals are probabilistic, not absolute.