JevForAgents
⌘K
中文Advertise
COMMUNITY OBSERVEDMetric claims: AUTHOR REPORTEDSOURCE AUDITED🤖 Jev Agent

jev-axi

A shell CLI for agents: pick, rate, check, rank, triage and guard.

🐙 Open Repository on GitHub ↗
MEDIA PREVIEW
Original jev-axi safety demo
Original jev-axi safety demo. Repository GIF demonstrates a blocked download-and-run command, a routine local allow and fetched-text screening.

Overview & result

Jev-axi is a command-line safety hook for coding agents. It asks Jev bounded hazard questions about risky Bash calls or fetched text, while ordinary commands may pass locally; the hook returns allow, deny or escalation according to policy.

From the original post

A second opinion for coding agents, in half a second.

— Nicholas Underwood

How Jev fits in the loop

  1. Intercept a planned tool call or untrusted fetched content.
  2. Apply local fast-path rules for routine operations; ask Jev to score hazards when judgment is needed.
  3. Return a policy verdict and explanation to the coding agent before the side effect occurs.

How to reproduce

  1. Install the documented CLI and register its `PreToolUse` hook in a test Claude Code or Codex setup.
  2. Compare a routine project command with the README's download-and-run example.
  3. Feed a known prompt-injection fixture through `jev-axi guard` and inspect exit code and human override path.

Why this build matters

Demonstrates guardrails at the actual side-effect boundary, including prompt-injection screening of tool output.

Limitations

  • Jev's probability is a judgment, not proof of danger or safety; local policy makes the final allow/deny decision.
  • The published GIF and examples do not establish live false-positive or miss rates.

Patterns

GuardrailsCoding Agents