Interactive Mode: Smart-Detect Strategy

Use VS Code's native ask_questions tool to gather structured user input through real UI widgets (dropdowns, multi-select, free text) when the request is ambiguous. Skip straight to work when intent is clear.

For the full ask_questions tool contract (schema, response format, hard constraints, and call patterns), apply the vscode-integration skill § askQuestions.

Core Decision Flowchart

code

┌─────────────────────────────────────────────────────────────────┐
│            SHOULD I USE INTERACTIVE QUESTIONS?                  │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Is the task type obvious from the request?                     │
│     NO  → Ask via ask_questions                                 │
│     YES → Infer and proceed                                     │
│                                                                 │
│  Is scope/complexity explicitly stated?                         │
│     NO  → Ask or infer from context                             │
│     YES → Use stated scope                                      │
│                                                                 │
│  Are focus areas mentioned or implied?                          │
│     NO  → Ask multi-select for priorities                       │
│     YES → Include mentioned areas + sensible defaults           │
│                                                                 │
│  Rule: If 2+ areas unclear → batch into one ask_questions call  │
│        If 1 unclear → infer default, note assumption explicitly │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Inference Heuristics

When keywords strongly signal intent, infer rather than ask:

Signal Words	Likely Intent	Confidence
"bug", "broken", "failing", "error"	Fix/Debug	High
"refactor", "improve", "clean up"	Refactoring	High
"add", "implement", "new", "create"	New feature	High
"should we", "compare", "evaluate", "which"	Decision/Analysis	High
"quick", "brief", "just"	Minimal scope	Medium
"thorough", "comprehensive", "deep"	Full scope	Medium

Rule: High confidence → proceed. Medium confidence → note assumption.

Interaction Design Rules

For detailed component patterns and trigger keywords, apply prompt-interaction-design skill.

Quick rules:

•Use ask_questions tool — never fall back to markdown-formatted questions
•Batch related questions (max 4 per call, 2–6 options each)
•Keep header ≤12 chars (it's both the UI label and the response key)
•Mark one option as recommended with a short justification in description
•multiSelect: true for additive choices; default (false) for either/or
•Summarize user choices in a table after interaction
•Don't re-ask unless requirements explicitly change

When to Trigger Interactions

Phase	Trigger Condition	Component Type
Start	Ambiguous scope (2+ unclear areas)	Multi-question `ask_questions` call
Options	2+ viable approaches exist	Single-select with trade-offs
Next Steps	Path forward unclear	Single-select for action

Anti-Patterns

•❌ Asking when intent is obvious from context
•❌ Using markdown-formatted questions instead of ask_questions tool
•❌ Asking one question at a time (batch into one ask_questions call)
•❌ Re-asking after user already clarified
•❌ More than 6 options per question (decision fatigue)
•❌ header longer than 12 characters (tool call will fail)
•❌ Missing "do nothing" option for optional changes
•❌ Using recommended on quiz/poll options (it pre-selects and reveals the answer)