agentleFS
NatchpaiNatchpai/Hardware-Copilot/.claude/skills/pcb-review/SKILL.md

pcb-review

Skill0 starsChanged 18 days agoMITView on GitHub

Reviews a KiCad PCB layout, read-only and evidence-based. Use when the user asks to assess layout risks, placement, routing, power, thermal, signal-integrity or manufacturability concerns in a PCB design. Not for looking up layout data only.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Start from the project's requirements, operating conditions, constraints and the user's review question.

    line 19
  2. Distinctive:

    Check the exact part, package, rating, pinout, thermal or layout guidance against the datasheet only when it changes a finding.

    line 42
  3. Distinctive:

    Use schematic connectivity or circuit intent when it is needed to read the PCB evidence.

    line 47
  4. Very specific:

    Never open or read a PDF. Datasheet values come from the user, official manufacturer web pages, or NotebookLM answers with cited text, used …

    line 57
  5. Very specific:

    If the project has .claude/rules/engineering.md, follow it.

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PCB Layout Review

Review the requested scope as an engineering assessment, not an approval or a design-state change. Native KiCad files are the source of truth. PCB, library, project and review files stay unchanged.

Scope and evidence

Start from the project's requirements, operating conditions, constraints and the user's review question. Pick the smallest useful focus: a board area, power domain, interface, component group, net or manufacturing risk. Choose checks from the design itself, not from a fixed circuit-type checklist or an architecture guessed from reference designators or net names.

Get targeted saved-board evidence with these queries (pcb-inspect if installed; otherwise targeted reads of the saved .kicad_pcb):

  • summary, footprint and nearby for placement context.
  • pads and pad-distance for transformed pad-centre geometry and assignments.
  • net for saved pad assignments only.
  • tracks and vias for saved routing primitive geometry and metadata.

Respect the bounded output, ambiguity results and NOT_VERIFIED states. A shared net name, track assignment, via assignment or geometric distance does not prove copper continuity, clearance, route completeness, return-path quality or electrical behavior.

Results reflect the saved .kicad_pcb file, not unsaved edits in the KiCad PCB Editor. Report editor-state consistency as NOT_VERIFIED unless it was checked.

Check the exact part, package, rating, pinout, thermal or layout guidance against the datasheet only when it changes a finding. Calculate only with a stated model, known inputs and assumptions. Explore alternatives only when design-level trade-offs need it.

Use schematic connectivity or circuit intent when it is needed to read the PCB evidence. PCB net assignments alone do not show the intended electrical behavior.

Evidence baseline

  • Label each significant claim VERIFIED (named evidence: CAD data, a calculation or an official source), ASSUMED (assumption stated), ESTIMATED (method and margin stated) or NOT_VERIFIED (evidence missing or check not run).
  • Never open or read a PDF. Datasheet values come from the user, official manufacturer web pages, or NotebookLM answers with cited text, used only after the user approves uploading the file. With none of these, the value is NOT_VERIFIED.
  • KiCad design and library files are read-only. Propose changes; do not apply them.
  • If the project has .claude/rules/engineering.md, follow it; it is the full policy.

Optional companions

  • pcb-inspect: targeted saved-board queries.
  • sch-inspect: schematic connectivity and circuit intent.
  • component-check: exact part, package, pinout and layout guidance.
  • eng-calc: calculations with a stated model.
  • design-assist: alternatives and design-level trade-offs.

These help if installed. If one is missing, do the step directly under the same evidence rules, or mark the dependent check NOT_VERIFIED. Call a companion only when the finding needs it.

Review reasoning

Assess only the topics that apply to the design and the evidence you have:

  • Placement: related parts together, connector access, separation of sensitive areas, orientation, and access for assembly.
  • Power and return paths: current loops, supply distribution, return-path breaks, decoupling placement and local energy storage.
  • Routing: layer changes, widths, routing geometry, reference-plane changes, differential or controlled-impedance constraints when specified, and coupling into sensitive nets.
  • Thermal and mechanical: heat sources, copper and thermal paths, component spacing, assembly, test and service.
  • Manufacturing: land-pattern evidence, solderability, polarity and orientation, tooling, keepouts, and stated fabrication or assembly limits.

Native KiCad DRC output and manual visual inspection are separate evidence sources. When either is available, report its exact source, scope and result; a missing DRC or visual check is not a pass. DRC does not prove electrical, thermal, EMC or manufacturing correctness. Ask for a targeted visual check when native data cannot settle geometry, layer relationships, filled-zone state or assembly detail.

Findings and actions

Give every conclusion a status:

  • Confirmed defect: direct CAD, DRC, datasheet, calculation or visual evidence shows a specific mismatch or violation.
  • Potential risk: the evidence supports a plausible concern but lacks the constraint, model or inspection to confirm it.
  • NOT_VERIFIED: the needed evidence is missing, ambiguous, unsupported or outside the tool's scope.

For each finding give: severity (CRITICAL: credible hazard or severe defect with clear evidence; MAJOR: likely functional or reliability issue; MINOR: bounded margin or documentation issue; INFO: observation, not a defect), the status above, the evidence label, target and location, evidence source, engineering impact, and one next check or corrective action. Severity rates the individual finding, not the board. Recommend changes; never apply them. Output sections: Scope, Findings, Evidence, Limitations / NOT_VERIFIED and Targeted next checks.

Example finding row:

MAJOR | C14 / U2 VIN | Potential risk | VERIFIED | C14 pad centre is 11.8 mm from U2 pin 3 (pad-distance); decoupling effectiveness unproven | pad-distance query, saved board | Higher supply loop inductance | Move C14 closer or confirm with a visual check

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.