菜单
Use when changing perf-sensitive codepaths (parsing, selector/extend algorithms). Provides a lightweight sanity protocol and avoids ungrounded perf claims.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Load BEFORE writing or changing code on Jess hot paths (core tree/eval/render, grammar/parser, extend/selector algorithms). Packages the 9 V8-architecture invariants as pre-write checks and points at the canonical checklist.
Use when working with fixture-based or expected-output tests (especially Less test-data). Helps isolate one fixture, reproduce deterministically, update expectations safely, and avoid cross-package build pitfalls.
Use when changing exports, entrypoints, or shared types across packages. Ensures public API stability, correct ESM/CJS surface, and cross-package build/test validation.
Keep reviews light by running quick spec and quality checks after each task before claiming completion.
Use a strict observe-hypothesize-trace-verify-fix loop when debugging; avoid random changes and context loss. Use when the user is debugging, investigating a failure, or fixing a bug.
Run and read the verification command every time before claiming success. Use when you are about to report a fix, tests/builds passing, or task completion.
| name | perf-sanity |
| description | Use when changing perf-sensitive codepaths (parsing, selector/extend algorithms). Provides a lightweight sanity protocol and avoids ungrounded perf claims. |
Use this skill when touching known perf-sensitive hotspots, especially:
packages/*-parser/src/** (tokenization/productions/parsing)packages/core/src/tree/util/extend.ts (extend selector algorithm)State the expected perf impact
Prefer “sanity checks” over claims
Keep changes locally contained
Watch for cross-package behavior
lib/ output confusion (build upstream first when needed).