Gptme Review

gptme:review

Delegate code review to gptme, which runs independent correctness, security, and test-coverage passes on your current changes.

When to Use

• User invokes /gptme:review
• User wants a second opinion or multi-lens review on the current diff
• Current changes are non-trivial and a systematic bug-finding pass adds value

When Not to Use

• Diff is tiny/mechanical (config change, typo fix)
• User just wants a quick summary, not a structured review
• gptme not in PATH

Procedure

# Review current staged+unstaged diff
DIFF=$(git diff HEAD)
if [ -z "$DIFF" ]; then
  DIFF=$(git diff --cached)
fi

if [ -z "$DIFF" ]; then
  echo "No changes to review (git diff HEAD is empty)"
  exit 0
fi

gptme --non-interactive \
  "Review the following code changes. Run three passes: (1) correctness — logic bugs, broken invariants; (2) security — unsafe input handling, secret leakage; (3) test coverage — missing regression tests, unverified edge cases. For each finding include: file, line, severity (high/medium/low), and a concrete fix hint. Output findings as a markdown list ordered by severity." \
  - \
  "$DIFF"

Review a specific file

gptme --non-interactive \
  "Review this file for bugs, security issues, and missing test coverage. Output findings as a markdown list ordered by severity." \
  - \
  "$(cat path/to/file.py)"

Tips

• The --non-interactive flag is required for non-blocking execution
• For large diffs, consider reviewing one file at a time
• gptme's review is independent of this Claude Code session — it won't share your prior context

Related

• gptme docs
• /gptme:run — delegate arbitrary tasks to gptme
• /gptme:context — inject gptme lessons into this session