fix: use local sparse clone in tool PR quality skill
loading diff…
The tool PR quality analysis skill was making hundreds of individual gh api calls per PR to read file contents. This was slow (~3-4 min per subagent), hit GitHub rate limits, and required pagination handling that was error-prone (silently missing files).
/tmp/Read tool on local files instead of gh api calls/tmp/ clonesEach sparse clone takes ~10-20 seconds vs ~3-4 minutes of API calls per subagent. For a 5-PR comparison, this saves ~15 minutes of wall time.
🤖 Generated with Claude Code