菜单
RES: Read-only audit of raw data: variable inventory, missingness, IDs, units, merge keys, feasibility. Produces data_audit_report.md. Trigger: audit data, inspect data, data audit, check raw data. Do NOT load for: cleaning, analysis, review, release, setup.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/maxwell2732/claudecode-research-harness-workflow --skill research-harness-audit命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | research-harness-audit |
| description | RES: Read-only audit of raw data: variable inventory, missingness, IDs, units, merge keys, feasibility. Produces data_audit_report.md. Trigger: audit data, inspect data, data audit, check raw data. Do NOT load for: cleaning, analysis, review, release, setup. |
| description-en | RES: Read-only audit of raw data: variable inventory, missingness, IDs, units, merge keys, feasibility. Produces data_audit_report.md. Trigger: audit data, inspect data, data audit, check raw data. Do NOT load for: cleaning, analysis, review, release, setup. |
| kind | workflow |
| purpose | Perform a read-only audit of raw data files and produce a structured audit report |
| trigger | audit data, inspect data, data audit, check raw data, /research-harness-audit |
| shape | evaluate |
| role | evaluator |
| pair | research-harness-clean |
| owner | research-harness-core |
| since | 2026-05-29 |
| allowed-tools | ["Read","Write","Bash","Glob","Grep"] |
| argument-hint | [--file PATH] [--all] |
| user-invocable | true |
| effort | medium |
Perform a read-only audit of raw data files. No data file is modified. The only output is a structured audit report and a log.
This skill runs after /research-harness-setup and before /research-harness-clean.
| Input | Action |
|---|---|
/research-harness-audit | Audit all files listed in study_spec.md §3 |
/research-harness-audit --file data/raw/X.csv | Audit a specific file |
/research-harness-audit --all | Audit everything found under data/raw/ |
Before starting:
study_spec.md. If it does not exist, stop and tell the user to run /research-harness-setup first.study_spec.md §3 exists. If it does not, report the missing path and stop.logs/audit_YYYYMMDD.log.For each file under data/raw/ (or the file specified by --file):
wc -l, head, column-sniffing) where possibleLog each file to logs/audit_YYYYMMDD.log.
For each file, record:
| Variable | Inferred type | Non-missing count | Missing count | Missing % | Min | Max | Sample values |
|---|
Use the actual variable names from the file headers. Do not rename or interpret variable names — record them as-is. If a variable name is ambiguous, note it in the audit report under §6 Open Issues; do not infer its meaning from the name alone.
For each file:
idhhidperson_idpid_id_codeDo not assume that two variables with similar names are the same ID. Report the candidate match and leave it as unknown if not confirmed by the data dictionary.
Identify candidate time variables (e.g., year, wave, date, month). For each:
high missingnessFor each file:
List pairs of files that appear to share a common ID variable. For each pair:
likely merge key, possible merge key, or unclearDo not perform any merge in this step.
Compare the raw data to study_spec.md:
Record each check as feasible, partially feasible, or infeasible with a one-line reason.
Infeasibility gate: If any required element is infeasible, the overall assessment is infeasible. Tell the user what is missing and that study_spec.md must be revised before cleaning can begin. Do not proceed to cleaning with an infeasible design.
Copy templates/data_audit_report.md to reports/data_audit_report.md and fill in all sections from Steps 1–9.
Save logs/audit_YYYYMMDD.log.
data/raw/reports/data_audit_report.md exists and all sections are populatedlogs/audit_YYYYMMDD.log existsdata/raw/ were modified (verify with git status data/raw/ or file-size check)feasible / infeasible verdictreports/data_audit_report.md existslogs/audit_YYYYMMDD.log existsTell the user:
Audit complete. Review
reports/data_audit_report.md.If the feasibility verdict is
infeasible: revisestudy_spec.mdbefore continuing.If
feasibleorpartially feasible: fill intemplates/data_cleaning_plan.md, save it asreports/data_cleaning_plan.md, review it, then run/research-harness-clean.