hooks — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited hooks (Hook) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Claude Code skills for hardware design, validation, and bring-up.
A plugin of focused skills that teach Claude how to do real hardware work: designing reusable HDL, integrating an SoC, bringing up FPGAs and bare-metal targets, building firmware boot chains, verifying designs against golden models, validating codegen on real silicon, prepping a tapeout, and packaging EDA tools in Nix. Each skill is domain-general but carries Midstall's house style: failures should fail, fix the root cause, test like it's going to silicon.
| Skill | Use when |
|---|---|
hdl-module-design | Writing or refactoring an HDL module and you need it parameterized, validated, and testable |
reference-driven-rohd | Implementing a block/protocol/peripheral in ROHD and you should fetch the spec, references, and reusable libs first instead of writing from memory |
rohd-rtl-gotchas | Building or testing RTL in ROHD/rohd_bridge/rohd_hcl and hitting hierarchy, sim-setup, or SV-emission errors |
soc-integration | Composing an SoC from peripherals and a bus, or generating device trees / ACPI from one description |
rtl-area-timing | Optimizing RTL microarchitecture for area or Fmax when a design won't fit or meet timing |
fpga-synthesis-fit | Synthesizing with yosys/nextpnr, fighting area/congestion, measuring Fmax, or instantiating block RAM |
fpga-bringup | Loading a bitstream onto a real FPGA and driving it over JTAG/GPIO |
bare-metal-bringup | Bringing up bare-metal or kernel code on a new arch/board that won't boot or hangs |
firmware-boot-chain | Building or debugging a firmware/boot chain (SBI/UEFI/ACPI/bootloader) or measured boot |
differential-verification | Verifying a DUT against a golden reference model with coverage-guided fuzzing |
codegen-validation | Proving a compiler backend's output by executing it on a real CPU or fast emulator |
tapeout-precheck | Prepping a design for tapeout: DRC/LVS signoff, submission packaging |
nix-eda-packaging | Packaging EDA toolchains in Nix without dlopen/wrapper hacks |
silicon-grade-discipline | Any silicon-bound code facing a fast-vs-correct tradeoff (the shared backbone) |
The plugin ships more than the skill files.
Agents (agents/) bundle the relevant skills into a focused reviewer you can dispatch at the right moment:
rtl-reviewer reviews an HDL diff against the design and discipline red flags.tapeout-signoff runs the precheck gate and refuses to wave real violations.bringup-debugger drives the bring-up checkpoint ladder when hardware is dead.Commands (commands/) are guided entry points that drive a workflow:
/fpga-fit runs the measure-don't-guess synthesis and timing loop./tapeout-check walks the tapeout signoff gate./hw-bringup drives the bring-up ladder for a stuck board.Hooks (hooks/) enforce the house style mechanically: a write-time check rejects em dashes, emoji, and malformed SKILL.md frontmatter, so the rules are automated rather than repeated in prose. The same logic runs in CI via scripts/lint-skills.sh.
Evals (evals/) are the benchmarks that prove the skills earn their keep: usefulness (do they beat the unaided model), discovery (do they fire at the right time), and adversarial discovery (do they resist bait and stay quiet on vague prompts). See evals/README.md. These skills are measured, not asserted.
Add the marketplace and install the plugin:
/plugin marketplace add Midstall/claude-for-hardware
/plugin install claude-for-hardwareSkills activate automatically when their trigger conditions match. You can also invoke one directly by name.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.