JevMade Sign in
← Back to guides

JevMade field notes / Technical guide

Check the evidence behind an AI assistant's claim to be done

jev-belay targets assistants that finish after editing files without running a meaningful check. Local rules inspect the work first; only an edited, unchecked turn is sent to Jev.

Original by valentynkitAgent workflows

Listen to this guide

JevMade’s plain-English explanation

0:00 /

Our summary

jev-belay addresses a narrow problem: an assistant says it is done after changing files without running a meaningful check. Local rules look for edits followed by a passing test, build, or code-style check. Most stopping points therefore need no model request.

Only an edited turn without a recognized check sends its task, final reply, change count, and check list to Jev. Four questions assess completion and whether checking applies. The program controls cutoffs, waiting limits, repeated decisions, and how often it may block a stop.

The approach misses false claims made without file edits and may misunderstand unfamiliar test output. Its reported results come from one private collection of conversations. Start by recording decisions without interruptions. A block asks for another check; it does not prove that check or the work is correct.

Key takeaways

  1. Inspect local work and test evidence before asking a model.
  2. Observe decisions before allowing interruptions.
  3. Limit blocks per request and session to avoid loops.

Performance and classification measurements come from the project author.

GitHub README

Read the original guide Opens the author’s site in a new tab.

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.