AI BriefingSafeAgents: Open-Source Proxy Intercepts Coding Agent Tool Calls Before Execution
·2026.10.29 21:05
Key point
SafeAgents is an open-source local proxy that uses up to three checks, including 110 deterministic detectors, to intercept and block malicious or out-of-scope tool calls before execution.
Details
- SafeAgents is an open-source local proxy that sits between a coding agent and its model server, intercepting tool calls before they reach the harness.
- The system performs up to three checks per call: verifying tool declaration, checking for undeclared hosts, and running atif-scan with approximately 110 deterministic detectors for sandbox escapes and reward hacking.
- Local checks execute in 10–20 ms on M-series Macs, while an optional judge model can be used for milder findings without exposing agent reasoning or tool outputs.
- The proxy supports OpenAI Chat Completions and Anthropic Messages APIs, with setup guides for 12 harnesses and end-to-end CI testing for six of them, including pi, Claude Code, and OpenHands.
- A --log-only mode is recommended for initial deployment to record attempts without altering trajectories, and the project is licensed under Apache 2.0.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
JetBrains releases ThinkRail, an open-source agentic IDE built on git worktrees and spec graphsNvidia OpenShell: Secure Runtime for Autonomous AI Agent Fleets