一键导入
evaluation-experiment-review
对实验结果、功能试运行和评审意见进行证据化判断与改进建议输出
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。
菜单
对实验结果、功能试运行和评审意见进行证据化判断与改进建议输出
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
Use when the user asks for a deep code review, adversarial review, multi-agent review, risk-ranked review, PR review, security-sensitive review, or implementation verification with concrete file and line evidence.
Use when the user asks to debug a failing test, runtime error, regression, flaky behavior, production-like failure, root-cause analysis, or loop-until-done repair workflow.
Use when the user asks for source-backed research, official-documentation lookup, technical comparison, dependency evaluation, repository investigation plus synthesis, or fan-out research with ranked conclusions.
Use when the user asks for a dynamic workflow, multi-agent workflow, task-specific harness, complex task orchestration, fan-out synthesis, adversarial verification, generate-and-filter exploration, tournament-style solution search, or loop-until-done execution in Codex.
Wireframe workflow for low-fidelity structure, hierarchy, and information layout exploration.
Animation workflow for motion-heavy artifacts, scene timing, and browser playback review.
| name | evaluation-experiment-review |
| description | 对实验结果、功能试运行和评审意见进行证据化判断与改进建议输出 |
| version | v0.1 |
| template_version | 1.1 |
| owner | xiaoz |
| cn_title | 评估与实验复核 |
| en_alias | evaluation-experiment-review |
| skill_family | evaluation |
| primary_artifact | evaluation verdict memo |
references/evaluation-patterns-sanitized.md:评估 memo 的判断结构与结论表达references/badcase-patterns-sanitized.md:bad-case 分类方式与表格模式references/source-coverage.md:本 skill 可使用的脱敏来源覆盖说明| Category | Typical tool types | Enable when | Avoid when | Expected evidence |
|---|---|---|---|---|
| Local docs/files | shell / file read / pdf/text review | 结果保存在本地文档、会议记录、反馈表里 | 没有结果材料时 | 样本摘录、反馈来源 |
| Structured docs | markdown / verdict memo / bad-case table | 需要形成 go/no-go memo、评审摘要 | 仅口头讨论时 | 结构化结论与行动项 |
| Data analysis | spreadsheet / metric comparison / segmentation | 有指标、分组结果、A/B 数据 | 没有统计材料时不要伪量化 | 指标表、分层观察 |
| Web research | official docs / benchmark references | 需要对照外部标准或最新平台规则 | 纯内部试运行复盘时 | 外部标准来源 |
| No-browse reasoning | direct synthesis only | 所有结果材料都在本地且不依赖最新事实 | 需要行业 benchmark 时 | 基于现有样本的判断 |
请帮我复核“当前页面问答”功能的灰度结果:我给你 20 条用户提问样本、人工标注的满意/不满意、以及 3 类失败案例。请输出这个版本是否适合扩灰,哪些问题是 blocker,哪些只需要快速修复。