メニュー
How to run a regression correctly - always "make clean-all && make run-all-{gate,func,full}-parallel", never a bare pytest. Use whenever running or re-running any test suite in val/ or projects/components/, or before reporting a suite as passing.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Authoring a per-module reference page under docs/markdown (RTLCommon/RTLMath/RTLAmba) - the local .md style (header table, Overview, Architecture, Top-level Interface, Data model, Timing, Related, Test) and the SV header block that mirrors it. Use whenever you add or change an RTL block and need its docs/markdown page, and keep the book index in sync.
What kind of documentation lives where in this repo - handbook (method) vs docs/markdown (reader-facing) vs vault/Tasks (work) vs beside-code (CLAUDE.md, PRD). The rule that a beside-code README is a link, never a standalone guide, and that style guides / how-tos never live in the RTL tree. Use before creating, moving, or "tidying" any .md file, and when reviewing an area's meta-docs.
Coverage methodology - Verilator line/toggle coverage per component, functional coverage expectations, and the monbus packet-type coverage matrix. Use when adding tests, closing coverage, or preparing a board-gate sign-off.
The Sherpa documentation pipeline - md_to_docx with --style, LoF/LoT/LoW caption encoding, book indexes, RTL PDF generation, HAS/MAS vs operator guides. Use before generating or restructuring any deliverable doc.
Filelist rules - EVERY module MUST have a .f and MUST be registered in bin/filelists.toml. Components own their compile closure, consumers -f include, never hand-list another area's sources. Use when adding ANY new RTL module, test, or .f file.
Formal verification flow - SymbiYosys via sv2v flatten, in-RTL ifdef FORMAL properties, mutation-checking every new assertion, harness vacuity traps. Use when writing properties, regenerating proofs, or after changing any module with a formal dir.
| name | regressions |
| description | How to run a regression correctly - always "make clean-all && make run-all-{gate,func,full}-parallel", never a bare pytest. Use whenever running or re-running any test suite in val/ or projects/components/, or before reporting a suite as passing. |
READ FIRST: vault/handbook/dv/running-regressions.md (the handbook is the repo's memory; this skill is the signpost).
The one rule agents break: clean-all is not optional. Stale
local_sim_build//sim_build//__pycache__ survive your edit and hand you a
green run against the OLD design. Skipping the clean does not error - it makes
the result more optimistic than the code deserves.
source env_python
cd val/amba && make clean-all && make run-all-full-parallel
Never substitute a bare pytest <dir> - that silently drops the level,
parallelism, and reruns the target supplies. Report pass COUNTS, not just rc=0.
The handbook root is vault/handbook/INDEX.md - design/, dv/, fpga/ areas, atomic notes, wikilinked. When you learn a durable lesson in this domain, ADD IT TO THE HANDBOOK NOTE, not to this skill.