菜单
Audits an ML training experiment for reproducibility, evidence integrity, leakage, checkpoint identity, evaluation validity, and overclaimed conclusions. Use before accepting results, spending on a longer run, publishing artifacts, or handing work to another agent.
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。
基于 SOC 职业分类
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/bastos/skills --skill audit-training-experiment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。
正在显示 SKILL.md
| name | audit-training-experiment |
| description | Audits an ML training experiment for reproducibility, evidence integrity, leakage, checkpoint identity, evaluation validity, and overclaimed conclusions. Use before accepting results, spending on a longer run, publishing artifacts, or handing work to another agent. |
Audit the claim against the preserved evidence. Do not regenerate missing evidence during the audit unless explicitly asked; identify the gap.
Verify that the record contains:
Confirm that preserved hashes match current artifacts. Distinguish unavailable evidence from failed evidence.
Check for:
Recompute inexpensive counts and hashes from preserved artifacts. Do not overwrite original reports when correcting them; create a clearly named addendum or new revision.
List each gate as passed, failed, not run, or not applicable with an evidence path. State the narrowest defensible conclusion and the smallest concrete next fix. Preserve negative evidence and prevent incompatible runs from sharing a state file or revision label.
Writes and revises clear, precise, accessible technical content for any medium. Use for manuals, wikis, README files, developer and user documentation, tutorials, references, specifications, RFCs, feature requests, issues, commit messages, code comments, review comments, pull request messages, release notes, runbooks, decision records, status updates, email, Slack, and other workplace or agent-facing communication.
Audits and designs lean, coherent context across system prompts, persistent instructions, repository guidance, skills, tool descriptions, memory, and references. Use for context engineering; creating, reviewing, simplifying, or reorganizing agent context; resolving conflicting, duplicated, bloated, stale, or misplaced guidance; designing progressive disclosure; or deciding where shared information should live.
Orchestrates and ships a milestone, release, issue set, or multi-workstream engineering goal through one manager and bounded parallel workers. Use when the user asks to manage a milestone, build a dependency graph, execute independent issues in parallel, coordinate threads, sessions, subagents, or worktrees, recover a stopped delivery, or invokes /gogopowermanager.