Show HN: Kintsugi – a local-first safety net for AI agents and humans
Category: security
Tags: ai-safety, command-line-security, rust
Score: 7.5/10 (Innovation: 8, Technical: 8, Documentation: 7, Utility: 7)
Kintsugi is a local-first safety net that intercepts shell commands from AI agents and humans, using a deterministic rule engine (not an LLM) to block catastrophic commands before execution, with tamper-evident logging and reversible undo. It combines AST-level parsing, hash-chained audit trails, and enterprise controls like password-locked settings and fail-closed operation, making it a novel approach to safe AI agent deployment.
Target audience: devops, security engineers, platform engineers
Repository: https://github.com/arrowassassin/kintsugi · Rust · MIT
View on Hacker News