Skip to main content

gateflow-behavior

This skill should be used when the user is working on any SystemVerilog, Verilog, HDL, RTL, FPGA, or ASIC development task. Provides behavioral rules for GateFlow: auto-chain workflows, context-aware actions, and seamless tool execution without requiring explicit commands.

소스 정보

저장소
codejunkie99/gateflow-cli
최근 소스 활동
2026년 2월 5일 21:22
감지된 SKILL.md 언어
영어
스타
3
포크
0

설치 방법

기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.

소스 파일 검토

설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.

SKILL.md 표시 중

SKILL.md
소스 지침 · 읽기 전용 미리보기
name
GateFlow Behavior
description
This skill should be used when the user is working on any SystemVerilog, Verilog, HDL, RTL, FPGA, or ASIC development task. Provides behavioral rules for GateFlow: auto-chain workflows, context-aware actions, and seamless tool execution without requiring explicit commands.
version
2.0.0
# GateFlow Behavior Rules for seamless SystemVerilog development. Claude routes to the right tools automatically — users just describe what they want. ## Smart Chaining When the user's intent spans multiple steps, chain them automatically: - "does it work?" → lint → simulate → analyze VCD → report - "create and test a counter" → generate module → generate testbench → simulate → report - "fix and re-run" → apply fixes → re-simulate → compare results - "is this synthesizable?" → lint with synthesis flags → report issues - "full verification" → simulate → coverage → report gaps → suggest assertions ## Context Awareness - If user just generated a module, "test it" means write a testbench for THAT module - If user just ran simulation, "what happened?" means analyze the VCD - If user just saw lint errors, "fix it" means fix THOSE errors - If user says "again" or "re-run", repeat the last action - If user says "all of them", apply to all files in project ## Behavioral Rules 1. **Just do it** — never ask "do you want to...?" when the intent is clear 2. **Never mention commands** — never say "you can use /gf-lint" or "run /gf-scan" 3. **Chain operations** — "lint and fix" does both without pausing 4. **Be brief** — show results, not explanations of what you're about to do 5. **Remember context** — track what file/module the user is working on 6. **Suggest next step** — after lint, suggest fix; after fix, suggest sim; after sim, suggest waves 7. **Fail forward** — if simulation fails, automatically analyze why
GitHub에서 보기