/audit-interaction

description: Evaluate an existing AI interaction against interaction mode taxonomies.
argument-hint: "[description of AI interaction to audit, or paste a conversation transcript]"

You are auditing an existing human-AI interaction. Use only skills from the model-interaction-design plugin. Follow this process:

Step 1: Classify the Interaction Mode

Using **conversation-patterns**, identify:

• What dialogue structure is being used? (interview, co-creation, instruction-execution, exploration, guided workflow)
• Is this the right structure for the task?
• Are there moments where the structure breaks down?

Step 2: Evaluate Turn-Taking

Using **conversation-patterns**:

• Are turns appropriately sized for the task?
• Does the AI know when to stop talking?
• Are there awkward turn boundaries?
• Rate turn-taking quality (1-5) with justification

Step 3: Assess Initiative Balance

Using **mixed-initiative-flow**:

• Who leads at each stage? Is this appropriate?
• Are handoffs clean or confusing?
• Does the AI take initiative when it should? Hold back when it should?
• Are there initiative anti-patterns (whiplash, passive AI, overbearing AI)?

Step 4: Check Repair Mechanisms

Using **conversation-patterns** and **feedback-loops**:

• When misunderstandings happen, how are they repaired?
• Can the user correct the AI easily?
• Does the AI acknowledge and recover from errors?
• Are there grounding checkpoints?

Step 5: Evaluate Disclosure

Using **progressive-disclosure**:

• Does the user know what the AI can do?
• Are capabilities revealed at the right pace?
• Is there evidence of underuse (user doesn't know about features)?

Step 6: Review Context Design

Using **context-window-design**:

• Does the AI maintain context across the conversation?
• Are there moments where context is lost?
• Is memory handled gracefully?

Output

Deliver an interaction audit report: 1. Overall quality score (1-5) with justification 2. Findings table: Issue | Severity | Skill Area | Recommendation 3. Top 3 improvements ranked by impact 4. Revised conversation flow showing recommended changes