grok-jev-guard
Grok Bot preflight layer: local hard gates first, then Jev judges ambiguity and returns proceed, scope_down, ask_human or block.
Overview
grok-jev-guard sits immediately before a meaningful Grok Bot tool sequence, receives a compact description of the pending operation and returns one explicit action: proceed, scope_down, checkpoint_then_run, change_approach, ask_human or block. Responsibility is split: local code owns secrets, deletion, publishing, payments, account changes, permissions and retry limits; Jev makes typed judgments about intent, approval expectation, destructive potential, scope fit and risk; Grok Bot executes within the returned limits and verifies; humans approve consequential or irreversible actions.
Unlike guards that trust a model's classification as policy, explicit high-consequence facts are evaluated locally before the Jev SDK is even imported, so a model answer cannot downgrade flags such as publishes_external, financial_action or deletes_data, permission or account changes, or secret-bearing state. State is redacted before Jev sees it. It is not a model router; it is an inspectable execution boundary.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What should happen with this pending operation? | proceed · scope_down · checkpoint_then_run · change_approach · ask_human · block |
| Yes / no | Does this operation have destructive potential / exceed the intended scope?(inferred) | yes · no |
How it works
- Describe the pending operation
- Local hard gate checks high-consequence facts
- Redact and ask Jev for typed judgments
- Return an action and limits; Grok Bot executes and verifies
Strengths
- Model cannot override hard local gates
- Redaction before model calls
- Graduated actions beyond allow/deny
Caveats
- Specific to Grok Bot workflows
- Requires a TypeSafe key
Getting started
git clone, python -m pip install -e ., copy config.example.yaml to config.yaml and set TYPESAFE_API_KEY.At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- preflight library / skill
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
\`grok-jev-guard\` sits immediately before a meaningful Grok Bot tool sequence. It receives a compact description of the pending operation and returns one explicit action:
logicrw/awesome-jev-projects
Frequently asked questions
Can Jev override a hard rule?
No, high-consequence facts are evaluated locally before Jev is called.
What actions can it return?
proceed, scope_down, checkpoint_then_run, change_approach, ask_human or block.
Is it a model router?
No, it is an execution boundary.
Related projects
Auditable Zig terminal agent pairing Codex for coding with Jev typed preflight and postflight checks under deterministic policy gates.
CLIproductionZigGateway that checks agent tool calls with policy plus Jev evidence, then binds approval to a single-use permit for that exact action.
MCP serverbetaTypeScriptMCP proxy and hooks that screen agent tool calls before they run and their results before the agent reads them, using Jev judgments.
MCP serverbetaTypeScriptXano backend template where Jev judges each agent tool call as safe, needs review or dangerous, and uncertain calls wait for a human.
ProjectbetaXanoScriptOpenAI-compatible endpoint that answers an agent's command-approval call with APPROVE, DENY or ESCALATE from a Jev Choice, failing closed.
ProxybetaPythonClaude Code plugin adding Jev judgments to Bash, Write and Edit calls and their outputs; a semantic guardrail, not a sandbox.
Pluginbeta