Canny
Coding-agent hooks use Jev to judge project rules and completion claims.
Added to Jevfast
How it uses Jev
Canny sends project rules and a code change to Jev as one Noul question per rule; a confident indication of a broken rule produces a note. At session stop, it can also ask whether the final message claims completion, which affects its check gate. Pattern checks for secrets and test damage are handled in code.
What Jev decides
Canny completion-claim checkExample answers · not a recorded Jev response · Source ↗
Question 1 · claims done
YOUR APP
INSTRUCTION
Does the final message claim the requested work is complete?
STATE
At session stop, the final message is evaluated only when code files changed and no passing verification was recorded.
JEV · NOUL
YesNo
App workflow
- Inspect tool activity
Record edits and commands in the session ledger; deterministic checks can block secret exposure, test removal, or repeated failures.
- Judge project rules
For changed files, send each project rule and the added or removed text to Jev as Noul questions; confident violations produce a note.
- Check completion
If code changed without a passing check, ask Jev whether the stop message claims completion, then apply the hook's check-gate policy.
Why it is interesting
It gives Jev narrow judgment calls inside a mostly deterministic hook: code owns the hard safety checks and gate, while Jev interprets project-specific rules and the meaning of a completion claim.