菜单
Compatibility alias for validated corpus completion. Use when a user says /sl-scale or asks to batch-label the remaining corpus with the legacy scale workflow; route to /sl-complete and require sealed-test qualification, risk routing, terminal reconciliation, final audit, and provenance.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/jluo41/Tools --skill sl-scale命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
PostFn specialist -- designs/reviews the response-formatting function in an Endpoint_Set. One of 5 inference Fn-types. Called by /haipipe-end when intent references PostFn, response formatting, post-processing, or `post`.
Run any Stage 6 endpoint work: routes intent to the right specialist across Fn-type, artifact-verb, develop-target, and deploy-target axes. Use for designing inference Fns, packaging Endpoint_Sets, training a build, local inference tests, or deploying anywhere. Trigger: endpoint, deploy, develop, train, package, inference Fn, MetaFn, TrigFn, PostFn, Src2InputFn, Input2SrcFn, /haipipe-end.
Src2InputFn specialist -- designs/reviews the record->wire-payload function in an Endpoint_Set (serializes a ProcessedDF record into JSON the model can ingest). Platform-specific: one impl per deploy platform (SageMaker flat JSON vs Databricks dataframe_records); --platform picks (default sagemaker). Called by /haipipe-end when intent references Src2InputFn, record-to-payload serialization, or `src2input`.
正在显示 SKILL.md
| name | sl-scale |
| description | Compatibility alias for validated corpus completion. Use when a user says /sl-scale or asks to batch-label the remaining corpus with the legacy scale workflow; route to /sl-complete and require sealed-test qualification, risk routing, terminal reconciliation, final audit, and provenance. |
Invoke /sl-complete with the same project path and compatible arguments.
Explain that scaling now requires a production route that passed sealed evaluation and ends with probability audit and provenance. Do not use unvalidated k-NN inheritance, classifier thresholds, or weak-model majority as a fallback.