JevMade field notes / Written guide

Add layered safeguards to a coding agent

Pi Jev Guard is an experimental extension that checks actions proposed by coding software. It combines exact local rules for known hazards with optional Jev questions about destructive work, unauthorized data transfer, task drift, and conflicts with project instructions.

Original by zszz3Guardrails

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits and license

“Pi Jev Guard” by zszz3. 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 Pi Jev Guard contributors

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

Pi Jev Guard checks actions proposed by a coding agent. Exact local rules catch known hazards first; optional model judgments about meaning look for destructive work, unauthorized data transfer, task drift, and conflicts with project instructions.

Rules can allow, warn, request one-time confirmation, or block. A credential is a secret used for access. Redaction replaces recognized secrets before online review, while non-interactive use means no person is present to answer a confirmation prompt.

The documented defaults do not specially block every upload of a credential file; without another matching rule, such an action may be allowed. Online checks can fail or be misled. This experimental, version-specific, unofficial extension is not a sandbox or complete permission boundary.

Key takeaways

Put known dangerous patterns in ordinary local rules.
Choose what happens when online checks fail or no person is available.
Secret removal and reminders are extra precautions, not complete protection.

More guides →See the experiments →