What I learned counting AI co-authored commits across 26,000 of my own

Description and use

The author discovered that 39% of his 26,251 commits involve AI co‑authors by leveraging an undocumented GitHub search qualifier with email addresses and corrected SVG rendering for GitHub’s image proxy.

Problem

Determining accurate counts of AI‑co‑authored commits in a large codebase was difficult due to undocumented GitHub search qualifiers, ambiguous co‑author trailers, and image rendering limitations.

Outcome

Accurate AI‑co‑authored commit counts (39% overall, 2% public) were achieved by using email‑based search qualifiers, validating against total commits, and correcting SVG rendering for GitHub’s image proxy.

Workflow

• 1. Run git log --format='%b' | grep -c 'Co-Authored-By: Claude' to get local counts.
• 2. Use GitHub CLI: gh api '/search/commits?q=author:USER+co-authored-by:[email protected]&per_page=1' --jq '.total_count' to query GitHub API.
• 3. Verify qualifier works by comparing counts for valid and invented qualifiers.
• 4. Check that total_count is less than total commits to ensure qualifier filtering.
• 5. Use email addresses from actual commit trailers (e.g., [email protected]) instead of display names to avoid false positives.
• 6. Build SVG card rendering counts, ensuring the SVG declares final geometry before animation to support GitHub’s camo proxy.
• 7. Test rendering in environments that ignore SMIL to confirm bars appear correctly.
• 8. Deploy the tool locally in CI so the user’s PAT can access private repos, avoiding the 2% public‑only limitation.

Evidence

10,261 of 26,251 commits — 39%.

Source

co-authored-by:[email protected] 10,261

Source

co-authored-by:cursor 10033042

Source