PR Learnt — Review Comment Teaching Loop

Scan PR review comments from the last 7 days across all your repositories and extract non-obvious lessons worth keeping. The goal is to make future sessions smarter, not to record every comment.

What's worth capturing

Look for:

• Corrections — reviewer requested changes: a pattern flagged, a wrong approach called out
• Validated approaches — reviewer approved with a substantive comment affirming the approach
• Decisions — a discussion in a PR that reached a conclusion about how to do something
• Your own review feedback — things you called out on others' PRs (they reveal your standards)

Ignore:

• Formatting/whitespace nitpicks
• Bot comments (Dependabot, CI bots, auto-reviewers)
• Trivial "LGTM" or "Looks good" with no body
• Things already in AGENTS.md, MEMORY.md, or global skill files

How to run