Show HN: Planlock – Approve an AI agent's plan once, code enforces every call

Category: security

Tags: ai-agents, mcp, security, guardrails, typescript

Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 6)

Planlock is a security guard for AI agents using MCP tools that enforces a human-approved plan at the code level, blocking any tool call outside the plan's defined limits. It provides per-step argument constraints, call limits, and version pinning to prevent prompt-injected or confused agents from overstepping. This approach offers a practical middle ground between per-call approval and fully unattended execution.

Target audience: AI engineers, backend developers, security engineers

Repository: https://github.com/JosephCurwin/planlock · TypeScript · MIT · 1 stars

View on Hacker News