Codex skill for maintaining a single append-only EXPERIMENT_LOG.md for research projects.
用于维护研究项目中单一、追加式 EXPERIMENT_LOG.md 的 Codex 技能。
This repository contains a standalone Codex skill named experiment-log-writing.
It helps Codex:
- create an
EXPERIMENT_LOG.mdheader, - append one run per section using a stable template,
- add summary or update blocks after batches of runs,
- track claim status for paper readiness,
- record failures, caveats, and supersession notes without rewriting history.
SKILL.md: the skill definition and workflow instructionsagents/openai.yaml: UI metadata for the skill
- Create a new experiment log for a benchmark project.
- Append a run entry from result files and commands.
- Summarize which claims are validated, pending, or superseded.
- Keep a paper-facing lab notebook consistent across runs.
这个仓库包含一个可独立使用的 Codex 技能:experiment-log-writing。
它用于帮助 Codex:
- 创建
EXPERIMENT_LOG.md文件头, - 按统一模板为每次实验追加独立条目,
- 在一批实验后追加总结或状态更新,
- 跟踪论文主张的验证状态,
- 在不改写历史记录的前提下记录失败、警告与结果替代关系。
SKILL.md:技能定义与工作流程说明agents/openai.yaml:技能的界面元数据
- 为新项目创建实验日志。
- 根据结果文件和运行命令补写一次实验记录。
- 总结哪些结论已经验证、待验证或已被替代。
- 维护适合论文写作与复盘的实验笔记。
Invoke the skill in Codex with a prompt such as:
Use $experiment-log-writing to append a new run entry to EXPERIMENT_LOG.md from results/latest.json.