菜单
Use when an experiment has completed and results need analysis, interpretation, or presentation. Triggers on "analyze results", "分析实验", "interpret experiment", "what did exp find", "run analysis".
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。
Unified document archival — covers both agent-facing knowhow (env setup, debug, infra, runbooks) and human-facing docs (design, guides, README, changelog). Triggers on: "update docs", "更新文档", "写 README", "记下来", "归档", "save this", "update knowhow", or after a <knowhow-hint> prompt. Also triggers when conversation context implies documentation should be created or updated.
Alias for /update-docs (knowhow branch). Use when user says '记下来', '归档', 'save this', or after a <knowhow-hint> prompt. Routes to the unified /update-docs workflow with knowhow branch auto-selected.
Use when creating git commits, updating CHANGELOG.md, committing across nested repos, or generating weekly progress summaries. Triggers on "commit", "提交", "changelog", "weekly summary", "周报", "write commit message".
Use when setting up a new project (general or research) or when .pipeline-state.json is missing. Triggers on "init project", "初始化项目", "set up labmate", "first time setup", "initialize".
Check experiment status and diagnose failures via exp-manager. Triggers on "check status", "看看实验", "how's the run", "any failures", "is it done", "monitor experiment", "experiment status".
Use when starting a new experiment — scaffolds exp/{exp_id}/ directory with README, config, run.py, analyze.py, and results structure. Triggers on "new experiment", "新实验", "scaffold exp", "set up experiment", "create experiment".
基于 SOC 职业分类
| name | analyze-experiment |
| description | Use when an experiment has completed and results need analysis, interpretation, or presentation. Triggers on "analyze results", "分析实验", "interpret experiment", "what did exp find", "run analysis". |
| disable-model-invocation | true |
Full analysis pipeline for the current experiment.
When this skill is invoked:
Read pipeline state:
import json
state = json.load(open('.pipeline-state.json'))
current_exp = state['current_exp']
If current_exp is null, ask user which experiment to analyze.
Run analysis script:
PYTHONPATH=. python exp/{current_exp}/analyze.py
This generates exp/{current_exp}/results/summary.md.
Delegate to @domain-expert for interpretation:
Tell it:
Analyze experiment {current_exp}: Read
exp/{current_exp}/results/summary.mdfor quantitative results. Readexp/{current_exp}/README.mdfor experiment context. Scandocs/papers/for relevant domain papers. Provide ~500 word domain interpretation with paper citations.
domain-expert has memory: project, so it remembers papers from previous sessions.
Merge interpretation into exp/{current_exp}/README.md under the "## Findings" section.
Update exp/summary.md cross-experiment table: update the row for current_exp with status "Analyzed" and key finding summary (one line).
Delegate to @slides-maker for presentation:
Tell it:
mode: analysis exp_id: {current_exp} Generate: slides/{current_exp}-analysis.html
slides-maker has background: true, so it runs in the background. Tell user: "slides-maker is generating slides in the background. You'll be notified when done."
Advance pipeline state:
stage to "analysis" in .pipeline-state.jsonPrint summary of generated artifacts:
exp/{current_exp}/results/summary.md — quantitative analysisexp/{current_exp}/README.md — updated with findingsslides/{current_exp}-analysis.html — presentation (generating in background)