菜单
Use when checking rebuilt Flat JS generated loader output, especially `example/opcode-kitchen-sink-loader.js`, for runtime-internal property names emitted as string properties, leaked bookkeeping fields, or cleanup regressions after runtime/compiler changes.
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。
Use when planning or implementing Flat JS VM state serialization, execution snapshots, snapshot restore, host capability registries, serializable runtime records, iterator/collection/function/generator/async serialization support, serialization playground work, or updates to `plan/vm-state-serialization-plan.md`. Guides phased work, runtime boundaries, unsupported-state policy, and validation gates.
Use when confirming a Flat JS branch is truly working, green, ready, or target complete after compiler, runtime, protected-mode, loader, example, or browser-facing changes. Provides the checklist and completion gates to satisfy before claiming the branch is done.
Use when fixing Flat JS bugs, regressions, unintended test failures, or behavior that broke after a recent change, especially in `src/compiler/**`, `src/runtime/**`, opcode work, or Test262-driven fixes. Enforces root-cause fixes: remove or narrow the regression-causing change first, and do not add workaround paths, new opcodes, special-case branches, sentinels, or duplicate semantics layers just to patch a newly introduced bug.
Use when adding or fixing ES2015+ language support in Flat JS, especially in `src/compiler/**`, `src/runtime/**`, opcode work, or syntax handlers. Enforces the project rule that newer JavaScript features must be implemented directly in compiler/runtime semantics rather than lowered into ES5-style helper code, synthetic AST rewrites, or userland compatibility shims.
Use when working on the Flat JS arena sim, debugger stepping, scan or lastMoveDistance barriers, TICKS_PER_SECOND semantics, or future multi-controller bot scheduling. Grounds changes in the current code so world ticks, raw VM steps, and user-visible highlight steps are not conflated.
基于 SOC 职业分类
| name | flat-js-loader-output-audit |
| description | Use when checking rebuilt Flat JS generated loader output, especially `example/opcode-kitchen-sink-loader.js`, for runtime-internal property names emitted as string properties, leaked bookkeeping fields, or cleanup regressions after runtime/compiler changes. |
Use this skill when the task is to verify generated Flat JS loader output, not just source files.
example/opcode-kitchen-sink-loader.js is one minified line and ignored by Git; line-based previews can hide misses.Rebuild the loader:
npm run build-example:opcode-kitchen-sink
Run the bundled audit script against the generated file:
node .codex\skills\flat-js-loader-output-audit\scripts\audit-loader-output.js example\opcode-kitchen-sink-loader.js
Directly inspect any reported suspect snippets in the generated output.
Fix real leaks at the source runtime/compiler path, not by editing ignored generated output.
Rebuild and rerun the audit until known internal leaks are gone.
Validate affected runtime behavior with focused tests, usually:
npm run build:tsc
npm test -- src/__tests__/es6-runtime.test.ts src/__tests__/scope-debug.test.ts
These should normally be removed from stripped loader output:
.names, .flags, .valuesnames:[], flags:[], values:[]__pos__Fields names such as "valueStack", "programSection", "variableEnvironment", "delegateIterator"A.pos = ... on thrown errors is diagnostic error metadata.{ _$_: run } is the CLI bootstrap binding for generated-loader execution.value and done are normal iterator-result surface names.name, length, prototype, constructor, get, set, return, throw, and next are normal JavaScript surface names.