FailproofAI/failproofai

Observability and enforcement for AI agent harnesses. Capture every run and runtime reliability with policy enforcement.

What it solves

FailproofAI provides observability and enforcement for AI agents. It prevents agents from executing dangerous tool calls—such as deleting files recursively, escalating privileges via sudo, or pushing directly to a master branch—by blocking them before they run. It also solves the problem of "black box" agent execution by providing a local dashboard and audit tools to trace model calls and tool usage.

How it works

Failproof hooks into 12 different agent harnesses, including coding CLIs (like Claude Code and GitHub Copilot CLI) and chat gateways (like Hermes and OpenClaw). It uses a policy-based system where each operation is evaluated against a set of rules. Policies can return one of three decisions: allow() to permit the action, deny(message) to block it and notify the agent, or instruct(message) to provide additional context for the agent's next prompt. For agents without a supported harness, a Python SDK is available for reporting and tracing.

Who it’s for

Developers and teams using AI coding agents or self-hosted assistants who need to ensure these agents operate within safe boundaries and maintain a full audit trail of their actions.

Highlights

  • Broad Harness Support: Hooks into 12 major agent environments, including Claude Code, Cursor Agent, and Devin CLI.
  • Pre-execution Blocking: Stops dangerous tool calls before they are executed, rather than just reporting them after the fact.
  • ** Built-in Guardrails**: Includes 40 built-in policies covering secret file protection, SQL safety, and infrastructure changes.
  • Local-First Observability: Features a local dashboard (localhost:8020) and offline audit tool that requires no account or external data transfer.
  • Customizable Policies: Allows users to write their own policies in JavaScript and commit them to a repository for team-wide enforcement.

Related

  • Project
    GoPlusSecurity/agentguard

    A real-time security layer for AI agents that blocks destructive commands, scans plugins for malicious code, and performs daily security audits to prevent data theft and system damage.

  • Project
    microsoft/agent-governance-toolkit

    A comprehensive governance framework for autonomous AI agents that provides deterministic policy enforcement, identity management, and audit logging to prevent unauthorized actions.

  • Project
    canyonroad/agentsh

    A secure, policy-enforced execution gateway for AI agents that intercepts and governs file, network, process, and database activity at runtime.

  • Project
    ucsandman/DashClaw

    A safety and governance layer for unattended AI agents that intercepts destructive tool calls and requires human approval via a signed decision ledger.

  • Project
    Justin0504/Aegis

    AEGIS is an open-source security firewall for AI agents that intercepts every tool call before execution, classifies it, checks it against policies, and either allows, blocks, or pauses it for human approval — with a dashboard, audit trail, and threat-coverage tracking.