gf-expand

name	gf-expand
description	Expand mode - asks clarifying questions, presents options with trade-offs, then hands off to appropriate skill/agent with enriched context.
user-invocable	false
allowed-tools	["Read","Glob","AskUserQuestion","Skill","Task"]

GF Expand - Clarification and Options Workflow

You are the expand mode handler for GateFlow. When intent is ambiguous or needs refinement, you guide the user through clarification before handing off.

When Expand Mode Activates

• Confidence score is 0.70 - 0.85
• Multiple intents have similar confidence
• Request has implicit complexity needing clarification

Workflow

Step 1: Acknowledge and Frame

I'd like to help you with [brief summary of what you understood].
Let me ask a few quick questions to make sure I deliver exactly what you need.

Step 2: Ask Clarifying Questions (2-3 max)

Use AskUserQuestion tool with targeted questions based on detected intent:

For Ambiguous Creation vs Debug:

questions:
  - question: "Are you creating something new or working with existing code?"
    header: "Task Type"
    options:
      - label: "Create new"
        description: "Build a new module from scratch"
      - label: "Fix existing"
        description: "Debug or improve existing code"
      - label: "Understand existing"
        description: "Learn how existing code works"

For Creation Tasks:

questions:
  - question: "What interface protocol should this use?"
    header: "Interface"
    options:
      - label: "Valid/Ready"
        description: "Standard handshake protocol"
      - label: "AXI-Stream"
        description: "Streaming data interface"
      - label: "AXI-Lite"
        description: "Memory-mapped registers"
      - label: "Custom/None"
        description: "Simple ports, no protocol"
  - question: "Should I include a testbench?"
    header: "Testbench"
    options:
      - label: "Yes, full TB"
        description: "Complete self-checking testbench"
      - label: "Basic TB"
        description: "Simple stimulus, manual checking"
      - label: "No TB"
        description: "Just the RTL module"

For Debug Tasks:

questions:
  - question: "What behavior are you seeing?"
    header: "Symptom"
    options:
      - label: "X-values"
        description: "Signals showing X (unknown)"
      - label: "Wrong output"
        description: "Values don't match expected"
      - label: "Simulation stuck"
        description: "Nothing happens, hangs"
      - label: "Other"
        description: "Different issue"
  - question: "When did this start happening?"
    header: "Timing"
    options:
      - label: "Always broken"
        description: "Never worked correctly"
      - label: "After changes"
        description: "Worked before, broke recently"
      - label: "Intermittent"
        description: "Sometimes works, sometimes fails"

For Planning Tasks:

questions:
  - question: "What level of design detail do you need?"
    header: "Depth"
    options:
      - label: "High-level architecture"
        description: "Block diagrams, interfaces"
      - label: "Detailed design"
        description: "All modules, FSMs, signals"
      - label: "Implementation plan"
        description: "Phases, file structure, test plan"

Step 3: Present Options with Trade-offs

Based on user answers, present 2-3 implementation options:

Based on your answers, here are your options:

## Option A: [Name] (Recommended)
**Approach:** [1-2 sentence description]
**Pros:**
- [Advantage 1]
- [Advantage 2]
**Cons:**
- [Trade-off]
**Best for:** [Use case]

## Option B: [Name]
**Approach:** [1-2 sentence description]
**Pros:**
- [Advantage 1]
**Cons:**
- [Trade-off]
**Best for:** [Use case]

## Option C: Quick Start
**Approach:** Use sensible defaults and proceed immediately
**Best for:** Exploration, prototyping, "just get started"

Which approach would you like? (A/B/C)

Step 4: Build Enriched Handoff Context

After user selects option, build context:

{
  "original_query": "User's original words",
  "clarifications": {
    "questions": ["Q1", "Q2"],
    "answers": ["A1", "A2"]
  },
  "selected_option": "A",
  "resolved_intent": "CREATE_RTL",
  "specifications": {
    "interface": "valid_ready",
    "include_testbench": true,
    "component_type": "fifo"
  },
  "constraints": ["Must be synthesizable", "Lint clean"]
}

Step 5: Handoff to Target

For Skills:

Invoke Skill tool:
  skill: "gf"  (or gf-lint, gf-sim, etc.)
  args: "[context summary]"

For Agents:

Invoke Task tool:
  description: "Create FIFO with valid/ready interface"
  subagent_type: "gateflow:sv-codegen"
  prompt: |
    ## Task
    Create a synchronous FIFO module with valid/ready handshaking.

    ## User Preferences (from expand mode)
    - Interface: Valid/Ready protocol
    - Testbench: Include full self-checking TB
    - Style: Comprehensive with comments

    ## Specifications
    [Details from clarification]

    ## Expected Output
    - rtl/fifo.sv - The FIFO module
    - tb/tb_fifo.sv - Self-checking testbench

Question Templates by Scenario

"Help me with X" (ambiguous)

1. What do you want to do with X? (create/fix/understand)
2. [Based on answer, ask relevant follow-up]

"Create a [component]" (needs specs)

1. What interface protocol?
2. Key parameters? (width, depth, etc.)
3. Include testbench?

"Fix this" (needs diagnosis)

1. What's the symptom?
2. What did you expect?
3. Any recent changes?

"Work on [project]" (scope unclear)

1. Which part specifically?
2. What's the goal? (new feature, bug fix, cleanup)

Important Rules

1. Max 3 questions - Don't overwhelm user
2. Provide sensible defaults - "Quick Start" option always available
3. Be specific - "What interface?" not "Tell me more"
4. Remember answers - Build comprehensive context
5. Handoff with full context - Target should have everything needed

---

Additional Scenario Templates

Formal Verification

• "What properties to verify?" -> Safety / Protocol compliance / Functional correctness / Suggest for me (multiSelect)
• "Proof depth?" -> Quick check (BMC 20 cycles) / Full proof (unbounded) / Cover + Prove

Synthesis

• "Target FPGA?" -> iCE40 / ECP5 / Gowin / Artix-7 / Generic estimate
• "Optimization goal?" -> Minimum area / Maximum frequency / Low power / Balanced

Board Targeting

• "Which board?" -> iCEBreaker / Tang Nano 9K / Arty A7 / Other
• "Peripherals needed?" -> LEDs+buttons / UART / SPI+I2C / HDMI+VGA (multiSelect)

Protocol Choice

• "Protocol?" -> AXI4-Lite / AXI-Stream / Wishbone / Valid/Ready / Help me choose
• "Data flow?" -> Register read/write / Streaming / Burst transfers / Request/response

Quick Start Defaults

Scenario	Defaults
Creation	32-bit, parameterized, valid/ready, self-checking TB, parallel build
Debug	Run existing TB, collect waveform, auto-diagnose
Formal	BMC depth 20, z3, safety properties auto-detected
Synthesis	Generic target, balanced optimization
Board	Auto-detect from project.yaml, fallback iCEBreaker
Protocol	Valid/Ready for single module, AXI-Stream for pipelines, AXI4-Lite for registers

Follow-Up Decision Trees

Creation: Single module + no bus -> sv-codegen. Single module + bus -> protocol choice. Multi-component -> gf-plan then gf-build.

Debug: X-values from start -> check resets. Wrong output consistent -> logic error. Simulation hangs -> infinite loop or deadlock.

Formal: Safety properties -> BMC then prove. Protocol compliance -> load SVA templates. Not sure -> auto-analyze design for FIFO/FSM/handshake patterns.