deep-plan（深度计划）

Deep Plan

Systematic approach for gathering requirements and clarifying decisions before implementation.

When to Activate

• User is in PLANNING mode
• User explicitly asks to "plan", "discuss", "let's think through", "review approach", "before we start", "what do you think about"
• Agent identifies choices between 2+ valid approaches
• Agent lacks context needed for confident decision-making

When NOT to Activate

• Fixing obvious bugs/typos (clear single correct solution)
• User explicitly says "do as you see fit" or similar
• Trivial changes with single logical approach
• User already provided specific requirements

Clarification Process

1. Context Questions First

Before technical decisions, understand the project context:

• Timeline/deadline constraints?
• Team size and experience level?
• Existing tech stack or constraints?
• Scale expectations?

These answers inform all subsequent recommendations.

2. Identify Decision Points

Scan the task and categorize decisions by priority:

• 🔴 Blocking — cannot proceed without answer
• 🟡 Important — affects outcome, but has reasonable default
• 🟢 Nice-to-know — can decide later

Types of decisions:

• Technical choices: libraries, tools, patterns, algorithms
• Architecture decisions: structure, component breakdown, data flow
• Scope boundaries: what's included vs excluded
• Priority conflicts: what to implement first
• Missing context: unclear requirements, ambiguous terminology

3. Progressive Questioning

Ask questions in phases, starting with most critical:

Phase 1: Context questions + 🔴 Blocking questions (max 5-7 per message) Phase 2: After blocking resolved, ask 🟡 Important questions Phase 3: Ask 🟢 Nice-to-know only when relevant phase begins

Do NOT dump all questions at once. Batch by priority and wait for responses.

4. Question Format with Confidence

Group questions by category, indicate recommendation confidence:

## [Category] Decisions

1. **[Brief description]** 🔴
   - Option A: [description] — **strongly recommend** because [reason]
   - Option B: [description] — better for [use case]

2. **[Another question]** 🟡
   - Option A: [description] — **lean towards** this
   - Option B: [description] — **no strong preference**, depends on [factor]

Confidence levels:

• strongly recommend — almost always the right choice
• lean towards — good option, but context-dependent
• no strong preference — both options are equally valid

5. Handle Unanswered Questions

If user does not answer some questions:

• Do NOT assume defaults silently
• Explicitly ask: "For questions X and Y, should I proceed with my recommendations, or do you have other preferences?"
• Wait for explicit confirmation before proceeding

6. Mid-Execution Re-check

If during implementation agent discovers:

• New ambiguity not covered by initial planning
• A decision that has multiple valid approaches
• Context that contradicts earlier assumptions

→ STOP and return to clarification mode. Do not make assumptions.

7. Decision Summary (Optional)

After all questions resolved, offer to create a decision log:

• "Would you like me to save these decisions to the implementation plan for reference?"
• Only create if user agrees
• Do NOT create files by default

Key Principles

1. Context before details — understand the big picture first
2. Ask first, implement after — when in doubt, ask
3. Progressive disclosure — blocking → important → nice-to-know
4. Explicit consent — never assume silence means agreement
5. Stay focused — max 5-7 questions per message
6. Re-check when uncertain — new ambiguity = new questions