review-stata

name: review-stata description: Run the stata-reviewer agent on a do-file. Produces a structured code-review report covering reproducibility, logging, naming, magic numbers, table/figure quality, and conformance to project conventions. disable-model-invocation: true argument-hint: "[path/to/file.do]" allowed-tools: ["Read", "Grep", "Glob", "Task"]

Review a Stata Do-File

Delegate to the stata-reviewer agent and report the findings.

Steps

1. Resolve target from $ARGUMENTS:

   • Full path: use it
   • Filename only: search dofiles/**/*.do
   • No argument: list dofiles/**/*.do and ask the user which to review

2. Confirm the file exists and is non-empty.

3. Invoke the agent:

   Delegate to stata-reviewer:
     "Review dofiles/<stage>/<file>.do.
      Read the do-file, the most recent log at logs/<stage>_<file>.log (if it exists),
      and follow stata-coding-conventions, stata-reproducibility-protocol, and quality-gates.
      Produce the report in the standard format."
   

4. Receive the agent's report at quality_reports/<dofile>_stata_review.md.

5. Summarize for the user:

   • Quality score
   • Total issues by severity
   • Top 3 critical/high issues with line references
   • Verdict: ready-to-commit / needs-revision

6. Offer to apply fixes for high-confidence single-line issues (per proofreading-protocol's propose-then-apply pattern). Never apply silently.

Examples

• /review-stata dofiles/03_analysis/main_regression.do → Full review with score and issue list.

• /review-stata main_regression → Searches dofiles/**/main_regression.do.

• /review-stata (no argument) → Lists candidates and asks.

Troubleshooting

• Agent reports "no log file" — run the do-file via /run-stata first; the reviewer benefits from log context.
• Score is high but you suspect a substantive issue — stata-reviewer covers code quality, not econometric correctness. Use econometric-reviewer (or /devils-advocate) for that.
• Agent disagrees with your style — adjust stata-coding-conventions.md for project-specific overrides; the agent reads that rule.

Notes

• This skill scores code quality, not result correctness. For result correctness use /validate-log (claims) and the econometric-reviewer agent (specs).
• The review report lives in quality_reports/ and persists across sessions; re-running this skill overwrites the previous report for that file.