requirement-parser

name	requirement-parser
description	Parse natural language hardware descriptions into structured JSON specifications, writing the result to workspace/<ModuleName>/spec.json.
disable-model-invocation	true
allowed-tools	["Read","Write","Bash","Grep","Glob"]

chip-agent:requirement-parser

Parse a natural language hardware description into a structured JSON specification file.

Usage

/chip-agent:requirement-parser <hardware description>

Instructions

1. Extract the hardware description from $ARGUMENTS. If $ARGUMENTS is empty or blank, ask the user: "Please describe the hardware module you want to create. For example: 'A 32-bit ALU supporting ADD, SUB, AND, OR, XOR, and SLT operations with a zero flag output.'"

2. Determine the project root directory by finding the nearest ancestor directory of this skill file that contains .claude. Use that directory as the base for all relative paths below.

3. Read the requirement-parser agent definition at agents/requirement-parser.md (relative to the project root). This agent contains the full parsing rules, schema definitions, and examples for both simple and complex hardware designs.

4. Follow the agent's instructions to parse $ARGUMENTS into a structured JSON specification. The agent returns ONLY the raw JSON -- no markdown fences, no explanation. Apply all parsing rules from the agent definition including:

   • Design complexity detection (simple vs complex)
   • PascalCase module names, camelCase port names
   • Correct port types (UInt, SInt, Bool) and widths
   • Operation expressions in Chisel syntax with io.<portName> prefix
   • SLT using signed comparison via .asSInt

5. Validate the JSON output before writing to disk:

   • moduleName is present and PascalCase (first character uppercase, alphanumeric only)
   • type is "simple" or "complex"
   • All ports have direction (input/output), type (UInt/SInt/Bool), and width (integer)
   • For simple specs: ports array is non-empty with at least one input and one output
   • For complex specs: every name in topModule.submodules matches a name in modules[]
   • If validation fails, fix the JSON before proceeding

6. Create the workspace directory structure using Bash:

   mkdir -p workspace/<moduleName>/generated workspace/<moduleName>/sim/logs
   

7. Write the JSON spec to disk using the Write tool. Write to:

   workspace/<moduleName>/spec.json
   

   This step is CRITICAL. The requirement-parser agent returns JSON to conversation output only -- it does NOT write files itself. This skill MUST capture the JSON and write it to disk so that downstream skills (testbench, compile, etc.) can read it.

8. Present a summary to the user:

   • Module name and description
   • Type: simple or complex
   • Port count: N inputs, M outputs
   • Operation count (if any)
   • Flag count (if any)
   • File location: workspace/<moduleName>/spec.json
   • Next step: "Run /chip-agent:testbench <ModuleName> to generate a testbench, or /chip-agent:compile <ModuleName> to compile to Verilog."

Input

$ARGUMENTS -- A natural language description of a hardware module (e.g., "A 32-bit ALU with ADD, SUB, AND, OR operations").

Output

workspace/<ModuleName>/spec.json containing the structured JSON specification. The workspace directory tree is also created with generated/ and sim/logs/ subdirectories.