jevbooks

← All projects

agent-chaperone

agent-chaperone/agent-chaperone · Homepage

Calibrated firewall screening AI agent tool calls and results.

What it is

agent-chaperone screens an AI agent's tool calls before they run and tool results before the agent reads them. It is an MCP proxy plus a hooks adapter for a client's built-in tools, for developers running MCP clients like Claude Code.

How it uses Jev

Jev makes the screening decisions: it takes the state of a tool call or result and returns calibrated probabilities for questions like destructive, exfiltration, severity, and instructs_reader. Those probabilities are compared against thresholds in a policy file, and the outcome is logged with the numbers that produced it.

Primitives:choicescorenoul

Technique worth stealing

Every decision comes back as a probability, every threshold lives in a policy file rather than a prompt, and every judgment is logged with the numbers that produced it.

Try it

npm install -g agent-chaperone, then wrap an MCP server or register hooks; starts in shadow mode.

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricno0.10
  • Shows a System One patternyes0.62
  • Handles uncertaintyno0.21
  • Measuredno0.12
  • Runnableyes0.96
  • Worth recommendingno0.09
  • Model replicano0.08
  • Problem scopescore on a 0–2 scale1.87
  • About Jevyes0.72

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.