Feat/workflow intelligence skills - #34
Conversation
Distil PCBWorld, pcbGPT, JLCPCB layout practice, and industrial review ops into reusable skills (architecture, layout/DFM/SI, power/clock, constraint SoT, design review, verification, product docs, agent ops) plus kidoc documentation generation — without duplicating analyzer engines. Additive Related Skills links from core kicad skill. Co-authored-by: Cursor <cursoragent@cursor.com>
Satisfy Phase 6 validation checklist section parity with peer methodology skills. Co-authored-by: Cursor <cursoragent@cursor.com>
|
Hi @TeinkBR — thanks for putting this together. There's real work here and I read all of it. As it stands I can't merge it, but I want to explain why rather than let it sit. kidoc: the PR restores the whole skill byte-for-byte from v1.3.2 — not sure if that was intentional. It was removed deliberately in v1.4 (report generation with a matplotlib/reportlab dependency stack conflicts with the project's structured-JSON, stdlib-only direction), and the analyzer output it consumes has changed a lot since then. I am taking it as a demand signal for documentation generation though — that's on the ideas list as a rebuild against the current analyzer contract. The methodology skills: there are ideas in here I like — non-claims sections, score ceilings without lab evidence, single-writer locks for agent CAD access all fit the direction of this project. But as written they're specific to your project's workflow (MatriQ, the EQ scorecards, the specific agent incidents), so they won't transfer to other users' boards, and 14 new trigger-bearing skills would compete with the existing kicad/emc/jlcpcb skills for routing while adding context weight for every plugin user. Skill descriptions are a tightly budgeted surface in this repo. If you want to rework it: drop kidoc, pick the single skill you think is strongest, make it project-agnostic, and keep its triggers from overlapping the existing skills. A small, focused version of this I can review properly. Thanks — it's clear you've been running agents against a real board, and that kind of field experience is exactly the feedback this project needs. |
|
Thanks a lot for the reply; And also sorry for the confusion from this commit, i think yesterday my cursor accidentally pushed the commit, I will rework it a bit later according to your suggestions. |
2 participants
No description provided.