菜单
Code review dispatching the Debugger agent in standalone mode (categories 4-12 only). Checks code quality, reproducibility, output standards, and professional polish. Use for R, Stata, Python, Julia, or any analysis script.
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
Sync the overleaf/ folder with the current content of paper/. Copies all LaTeX source, figures, tables, appendix, and bibliography. Puts the model section as overleaf/model.tex (top-level) for easy coauthor access. Run whenever paper/ changes and you want to push to Overleaf.
RA onboarding audit — explore all project folders and code, inspect data files directly, and produce a structured project-state document that enables a clean restart. Use at the start of a project hand-off or when picking up work from multiple prior contributors.
Full bibliography audit for an academic paper. Verifies every citation is (1) structurally correct in the .bib file, (2) actually cited in the paper, and (3) a real publication — author names, year, journal, and title verified via web search. Flags hallucinated or inaccurate references.
Design identification strategy by dispatching the causal-strategist agent (proposer) and identification-critic agent (critic). Produces strategy memo, pseudo-code, robustness plan, and falsification tests. Use when asked to "identify the effect", "design the strategy", or "write the strategy memo".
Show current context status and session health. Use to check how much context has been used, whether auto-compact is approaching, and what state will be preserved.
Show the current research pipeline status for the active project, or display the pipeline guide for a specific project type. Reads the research spec, research journal, and active plans to determine phase, scores, and next steps.
| name | review-code |
| description | Code review dispatching the Debugger agent in standalone mode (categories 4-12 only). Checks code quality, reproducibility, output standards, and professional polish. Use for R, Stata, Python, Julia, or any analysis script. |
| argument-hint | [filename or 'all'] |
| allowed-tools | ["Read","Grep","Glob","Write","Task"] |
Run the code review protocol by dispatching the Debugger agent in standalone mode.
In standalone mode, the Debugger runs categories 4-12 only (code quality). Categories 1-3 (strategic alignment) require a strategy memo and are only run within the pipeline or via /econometrics-check.
$ARGUMENTS is a specific file: review that file only$ARGUMENTS is all: review all scripts in code/, scripts/, or subdirectories by language$ARGUMENTS is a directory: review all scripts in that directoryFor each script (or batch), delegate to the debugger agent via Task tool:
Prompt: Review [file] in standalone mode (categories 4-12 only).
Adapt all checks to the language of the script (R, Python, Stata, Julia, etc.).
Categories:
4. Script structure (header, sections, flow)
5. Output hygiene (no debug print/log pollution, clean console output)
6. Reproducibility (random seeds, relative paths, no hardcoded values)
7. Function design (DRY, appropriate abstraction level, naming conventions)
8. Figure/output quality (labels, dimensions, consistent style)
9. Artifact saving pattern (all computed objects saved to appropriate output files)
10. Comments (explain why, not what)
11. Error handling (graceful failures, informative messages)
12. Polish (consistent style, no dead code, clean namespace)
Save report to quality_reports/[script_name]_code_review.md
After all reviews complete:
Do NOT edit any source files. Only produce reports. Fixes are applied after user review, either manually or by re-dispatching Coder (main Claude).