uygnoey Making Claude a better electrical engineer (skill)
Purpose
This skill converts the blog post's ideas into a reusable checklist for either (a) domain-expert partnership benchmarking, or (b) plan-fit assessment, depending on the post.
Instructions
Use this skill when you need to improve model performance on a specialized, tool-using workflow by partnering with domain experts.
- Define the target agentic task as an input → output transformation.
- Build an evaluation harness (testbench) that can grade outputs in domain terms.
- Prefer requirement-based checks over brittle exact component matches.
- Collect expert feedback to identify common failure modes and update the benchmark.
- Run blinded comparisons against baseline models to validate improvements.
When documenting the harness, keep the high-level requirements in the harness and reserve detailed rationale for references.
See: references/benchmark-notes.md.
Examples
- Goal: Improve a model that generates reference designs from chip documentation.
- Harness requirement: 'At least 22uF of capacitance between power and ground.'
- Evaluation: Run the testbench and compare outputs across candidate models.