Show & tell: open-source verify-before-stop hook for Claude Code

[ianymu]

Hey! Sharing an open-source Stop hook in case it's complementary to what you're building.

ianymu/claude-verify-before-stop — blocks Claude Code's Stop event if files changed but no VERIFY_ACTION + VERIFIED log entry exists from the last 5 min. Catches the very common failure mode where the model says "all tests passing ✓" without actually verifying.

Built over 12 months running Claude Code on 14 parallel projects. Pure bash + python3 stdlib, MIT, ~50 lines. Lives in .claude/hooks/verify-before-stop.sh and .claude/state/stop-verify.log for audit trail.

A few related artifacts you might find useful:

• Claude Code Hooks cheat sheet — Stop/PreToolUse/PostToolUse contract + 8-block cap gotcha
• Hook generator tool — customize the script for your stack
• Settings.json checker — audit your existing config for missing guards

Curious if anyone here has built related guards. Drop them in a comment and I'll add to the repo with credit.

— Ian

[ishita-0301]

This is a neat guard. One thing I'd consider adding is progress-based checks alongside verification. For example, if Claude keeps retrying the same command or enters a loop before reaching the Stop hook, fail early instead of waiting until the end.

I recently came across https://github.com/FailproofAI/failproofai, which explores similar runtime guardrails like loop detection and execution policies. Feels like these kinds of hooks could work really well together.