Flight recorder and firewall for AI agents — records every tool call, tracks lethal-trifecta taint across a session, and blocks the prompt-injection exfiltration path. Zero deps, cross-harness.
Stop AI agents from leaking your data or acting for an attacker. Information-flow control for Claude Code, MCP apps and Python agents.
Proves what cloud credentials, and the AI agents holding them, can reach. Z3-backed AWS IAM reasoning plus lethal-trifecta analysis for AI agents.
Find and block the lethal trifecta that forms across your MCP servers.
A hands-on lab and cross-vendor benchmark showing how indirect prompt injection makes an AI pricing agent leak a company secret — and why prompt guardrails aren't enough.
Flight mode for AI agents: Meta's Rule of Two enforced in-process, before the call. Benchmarked on AgentDojo.
Every skill you install says trust me bro. This one reads it first. Pre-install audit for agent skills, plugins, and MCP servers: lethal-trifecta reach, injection wording, provenance, and drift after approval.
Zero-dependency linter that flags the 'lethal trifecta' (private-data + untrusted-input + external-comm) in an AI agent's tool manifest — MCP, OpenAI, LangChain. CI gate. Concept: Simon Willison.
own your agent browser — an indirect-prompt-injection firewall + action gate for any CDP browser
To associate your repository with the lethal-trifecta topic, visit your repo's landing page and select "manage topics."