spice

SPICE simulation

This skill is independent of an EDA editor and MCP. Requires Node.js >=20.19. The scripts run directly as JavaScript; no compilation or Python is needed. Resolve script paths relative to this SKILL.md, not the current working directory.

Workflow

1. Obtain component values and actual net connections from the caller's schematic tools, supplied netlist, or circuit description. State the question being tested, supplies, input stimulus, load, and operating conditions. Detected circuit blocks are optional context, not a restriction on what can be simulated.
2. Find device models with node scripts/search.mjs "MPN". On first use it prepares two cache files: library.jsonl and index.json with byte offsets. It returns matching models, their text, exact modelId, original sourcePath, matching reasons, and validation scope. sourcePath is an archive identifier, not an existing filesystem path. Use --include-review only when ordinary results are insufficient. Similar names do not establish an equivalent component.
3. Save the selected model with node scripts/copy-model.mjs "<modelId from search>" --out ./simulation/models. This reads only that JSONL record and writes one .lib, returning its absolute path, model name and .include directive. Existing identical copies are reused; edited files are never overwritten. Read the complete saved model, including any text truncated in search. Establish port order from comments and the .SUBCKT declaration. Internal structure can support an inference but does not establish physical package pin numbers. Record inferred assignments; prefer another clearly documented model if important ports remain uncertain. No external pin-mapping table is required.
4. Write a standard .cir file with a title on its first line, .include/.lib declarations, sources, .save, analysis directives, and .end. Use .op, .dc, .ac, and .tran as needed. The runner owns .control; do not include a control block. Use separate circuits for repeated parameter cases. Include only selected model files to avoid name collisions.
5. Run node scripts/simulate.mjs circuit.cir --out results/run-01. Output directory must be new. The runner finds/prepares ngspice, sets ngbehavior=ltpsa before reading the netlist, runs with -n -b, exports ASCII raw/CSV/JSON and creates PNGs through Sharp. Numerical settings belong in .options in the netlist, not CLI flags. Relative includes resolve from each original file's directory and are copied recursively into the result's models/ directory.
6. Read result.json, measurements in ngspice.log, numerical data, and the PNGs. Check operating points, expected magnitudes, loading, saturation, and timestep/frequency resolution. Successful execution is not proof of model fidelity or correct circuit behavior. Compare important results with an independent calculation or a meaningful known limit.
7. Report what was simulated, the selected models and assumptions, quantitative results, and material limitations. Keep the complete result directory with the user's research: it contains the rewritten circuit and referenced model snapshots, with original paths and hashes recorded in result.json. External stimulus data files (e.g. file-backed sources) are not bundled; use inline sources or copy such data into the run explicitly.