Imported from oniondas/duck-rtl (
skills/duck-rtl/SKILL.md). Install upstream withnpx skills add oniondas/duck-rtl --skill duck-rtl. Copyright stays with the author.
Duck RTL — Verilog build/explore + token-optimized co-sim loop
Default behavior: write Verilog or explore the existing code in the repo. This skill provides deterministic guardrails for building and verifying RTL against a Python Golden Model, plus an optional AST-based FSM extractor (structure from the AST, labels from you — no LLM backend, no API key) for when a state diagram is explicitly requested.
When to use
- Write / scaffold a new Verilog module
- Understand, review, or modify existing RTL in this repo
- Verify a design against a Python golden model (ledger → scaffold → check → cocotb sim)
- Only when explicitly asked: diagram / visualize the control FSM of a module
Operating instructions
The full flow — the relabeling contract, the co-sim build loop (Rules 1–5), and the
Verilog-2001 constraints — lives in ${CLAUDE_PLUGIN_ROOT}/AGENTS.md. Read it before
proceeding; it's the single source of truth and is kept in sync for every agent.
Invoke the CLI through the plugin root so it works from any project directory
(${CLAUDE_PLUGIN_ROOT} is set by Claude Code to this plugin's install path).
Quick reference (FSM diagram, the common case):
python "${CLAUDE_PLUGIN_ROOT}/main.py" extract --file module.v --out graph.json # raw graph, deterministic
# read graph.json; write labels.json: {"states":[{"id","label"}],"edges":[{"id","label"}]}
python "${CLAUDE_PLUGIN_ROOT}/main.py" render --graph graph.json --labels labels.json --out fsm.svg
python "${CLAUDE_PLUGIN_ROOT}/main.py" run --file module.v --out fsm.svg # one-shot, raw conditions as labelsPrereqs (once): pip install -r "${CLAUDE_PLUGIN_ROOT}/requirements.txt", plus iverilog and Graphviz dot on PATH.