gf-map.md

name	gf-map
description	Map codebase
allowed-tools	Glob Read Write Task Bash Grep

GateFlow Map Command

Generate a comprehensive map of a SystemVerilog codebase with parallel analysis agents.

Instructions

1. Find all SystemVerilog files in the project:

   Use Glob to find **/*.sv and **/*.svh files
   

2. Create the output directory:

   mkdir -p .gateflow/map/modules

3. Spawn the gf-architect agent using the Task tool:

   • Pass the list of discovered files
   • The agent will spawn 10 sub-agents in parallel for analysis
   • Results are written to .gateflow/map/

4. Report completion with summary:

   • Number of modules found
   • Number of packages
   • Top module(s) identified
   • Any warnings (missing files, parse errors)

Output Structure

.gateflow/map/
├── CODEBASE.md          # AI-friendly summary (AGENTS.md style)
├── hierarchy.md         # Module tree with Mermaid
├── signals.md           # Signal flow diagrams
├── fsm.md              # State machine diagrams
├── clock-domains.md    # CDC analysis
├── packages.md         # Package dependencies
├── types.md            # Structs, unions, typedefs
├── functions.md        # Functions and tasks
├── macros.md           # Preprocessor directives
├── verification.md     # SVA, coverage, assertions
├── recipe.md           # Filelists, compile order
├── interfaces.md       # Interfaces, modports, clocking (if found)
├── classes.md          # Classes, UVM hierarchy (if found)
├── generate.md         # Generate blocks (if found)
├── dpi.md              # DPI imports/exports (if found)
└── modules/            # Per-module detail pages
    └── <module_name>.md

Analysis Method

Uses hybrid approach:

1. Token budgeting - Count tokens per file, group into ~150k chunks
2. Parallel agents - Spawn 2-10 agents based on codebase size
3. Regex parsing - Fast structured extraction of ports, instances, types
4. Merge - Combine agent context + regex structure

Incremental Updates

On subsequent runs:

• Detects existing .gateflow/map/CODEBASE.md
• Checks git history for changes since last scan
• Only re-analyzes modified files
• Merges updates with existing map
• Saves commit hash for next incremental run

First run: Full scan (may use significant tokens for large codebases) Later runs: Fast incremental update (only changed files)