Show HN: NILScript – OpenAPI for agent actions, so hallucinations can't write

Category: infrastructure

Tags: ai-agents, safety, openapi

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

NILScript provides a safety layer for AI agent actions by enforcing a propose-approve-commit-rollback workflow, preventing unauthorized writes even under prompt injection attacks. Its innovative approach treats agent actions like OpenAPI for agents, with structural guarantees against hallucinated or hijacked operations. The project includes benchmarks showing zero unauthorized writes across thousands of attacks, making it interesting for production agent deployments.

Target audience: backend devs, devops, ai engineers

Repository: https://nilscript.org/ · Python · NOASSERTION

View on Hacker News