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 /

AI narration

Credits and license

“jev-belay” by valentynkit. Read the original source.

This expanded guide is an AI-narrated adaptation of the source’s essential explanation, examples and caveats, not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

License: MIT

MIT License

Copyright (c) 2026 Valentyn Kit

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

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

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

More guides →See the experiments →