evidence-convention

仓库创建 2026年7月6日最近提交 21 天前SkillHot 收录 20 天前

▸ 精选理由

确保每个实验结果都有可审计的记录与meta，提升复现性。

这个 Skill 做什么

规定结果保存位置与格式，禁止“暗地运行”，要求生成元数据。

做需要保留或会被引用的计算实验时，把结果有章法地落盘并写好元数据（比如 meta.json），防止所谓的“暗地运行”（只存在缓存、scratch 或聊天记录）。它会规定证据要放在哪个目录和每次运行必须记录哪些信息，把元数据和精简摘要作为仓库要跟踪的记录，而不是把大体量原始数据塞进版本库。特别之处是把可复现性和审计变成硬规则，方便引用、复查与责任追踪。

▸ 展开 SKILL.md 英文原文

Use when launching any computational campaign whose results will be kept, cited, or decided upon — sweeps, batch runs, Monte Carlo draws, precomputes — and when auditing for "dark runs" (results that exist only in caches, scratch dirs, or chat logs). Defines where evidence lands and what a run must record. Trigger phrases: "where do the results go", "record this run", "write the meta file", "dark run", "promote these results", "run report".

研究检索证据管理实验复现数据组织通用

Stars

Forks

仓库内 Skill

7 日增星

安装 / 使用

给你的 Agent 一句话（通用）

帮我安装这个 skill：https://raw.githubusercontent.com/huguryildiz/ALETHEIA/main/skills/core/evidence-convention/SKILL.md

或 curl 直取 SKILL.md

curl -fsSL "https://raw.githubusercontent.com/huguryildiz/ALETHEIA/main/skills/core/evidence-convention/SKILL.md"

SKILL.MD 节选查看完整文件 ↗

# evidence-convention

**No dark runs.** A result that exists only in a gitignored cache, a scratch directory, a
terminal scrollback, or a chat transcript is not evidence — it is a rumor with numbers. The
contract: **every kept campaign lands in `{{evidence_dir}}/<name>_<YYYY-MM-DD>/` with a
`meta.json`, and what the repo tracks is the record (meta + curated summaries), not the
bulk.**

## When to use

- Launching a sweep, batch run, sampling campaign, or precompute whose output will be kept.
- A decision, build-log entry, or paper claim is about to cite a run — does the run have an
  evidence directory?
- Auditing the project for dark runs.
- Writing the run report after a campaign finishes

via SKILL·HOT · 数据来自 GitHub 公开信息 · 原文版权归作者所有