Show & Tell: How I caught 287 AI rule violations across 50 repos with SpecLock #7
sgroy10
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Show & Tell: How I caught 287 AI rule violations across 50 repos with SpecLock
I ran SpecLock's semantic audit across 50 active repos in my workspace and what I found was... humbling.
Total violations caught: 287 across repos where I thought my
CLAUDE.mdrules were being respected by Claude, Cursor, and Copilot. They weren't.Full writeup with charts and repo-by-repo breakdown here: https://sgroy10.github.io/speclock/blog/50-repos-claude-md-violations.html
The 7 drift patterns I saw repeatedly
CLAUDE.mdsaid "don't remove the auth middleware." AI "refactored it out." Same outcome, different word.CLAUDE.mdsaid "minimal diffs only." Ignored.CLAUDE.mdpinned a specific pattern (e.g., "use repository pattern for DB access"). AI quietly introduced direct ORM calls.Why keyword matching misses these
Most naive hooks do
grep -i "don't delete"and call it a day. That catches maybe 30% of violations. The rest hide behind synonyms, rewordings, and indirect actions. SpecLock's semantic engine uses a synonym taxonomy + intent detection to catch the other 70%.Your turn
Have you run a semantic audit on your repos yet? Drop the numbers in a reply:
How many violations did you find? Which drift pattern hit you hardest? Would love to crowdsource a drift pattern atlas from this community.
Related reading:
Beta Was this translation helpful? Give feedback.
All reactions