Failproof
Observability and policy enforcement for AI agent harnesses, hooking coding CLIs and chat gateways to capture every run and block dangerous tool calls before they execute.
Failproof hooks twelve agent harnesses, ten coding CLIs plus chat and assistant gateways such as Hermes and OpenClaw, under one policy API and one session history. Blocking a tool call before it runs is verified across all twelve harnesses, and turn-end gates work on eight of them; agents running in an unsupported harness can report through a Python SDK for tracing, sessions and audits, though enforcement there needs a hook added to that runtime.
Forty built-in policies can block reads of .env and other secret files, warn on repeated tool-call loops, block privilege escalation and destructive SQL such as unbounded deletes, block unreviewed Terraform or kubectl changes, block recursive file deletion, and block force-pushes or pushes to a protected branch. One policy that blocks commands aimed at Failproof itself is always on and cannot be disabled, since an agent that could pause it could also disable every other policy.
It installs with npm install -g failproofai, after which failproofai config wires up agents and a local daemon and a dashboard runs on localhost:8020. It is written in TypeScript.