菜单
Full pipeline rebuild from scratch to verify reproducibility. Clears the cache, re-runs every step, verifies outputs match what was previously committed. Requires a clean git state before starting, if git is used.
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。
Adversarial review of the most recent piece of work in the current session, delegated to a sub-agent with fresh context. Probes for errors, untested assumptions, and glossed-over decisions. Produces a short critical document for the user. Use when asked to "critique", "be skeptical", "challenge me", or "find what's wrong".
Keep the analytical pipeline hygienic, clear, and efficient. Propose changes before implementing. Limit changes to what is clearly needed -- do not over-engineer. Tasks (in priority order): testing, linting, refactoring, documentation, efficiency, sub-agent review. Requires a clean git state before starting.
Adversarial pre-submission peer review of a manuscript, calibrated to the standards of a specific journal type (general-interest PS, comparative, methods, general science, economics). Use when asked for a review or referee report.
R style guide: directory structure, .qmd runner, targets, naming, function docs. Use when writing or refactoring R code, or reviewing a diff for style consistency.
基于 SOC 职业分类
| name | sweep |
| description | Full pipeline rebuild from scratch to verify reproducibility. Clears the cache, re-runs every step, verifies outputs match what was previously committed. Requires a clean git state before starting, if git is used. |
Verify the repo is reproducible from scratch. If unsure at any step, STOP AND ASK.
Check git state. Run git status. If there are uncommitted changes, or the branch is not up to date with the remote, STOP and report back. If the project is not tracked with git, you can ignore this.
Identify pipeline outputs. List the folders that house pipeline-generated files ("output folders") and the files inside them ("output files").
Verify safety. Confirm that (a) all files in the output folders are pipeline-generated, and (b) all pipeline-generated files live inside the output folders. If either is false, STOP and report back.
Move outputs aside. Move every file in the output folders to a temp folder inside the project. Output folders should now be empty.
Clear the cache and rebuild. Delete the _targets/ store (or equivalent). Re-run the pipeline from scratch.
Verify the rebuild.
Recompile the paper if any table or figure was regenerated.
Report. One status report at the end: what was regenerated, any deltas vs. the originals, any anomalies.
