PromptSign – Sigstore signing and verification for AI instruction files

PromptSign is a public key infrastructure for AI instruction files. It signs and verifies skills, agents, and CLAUDE.md / AGENTS.md files that AI systems run. The site states: "Sign the prompts your agents obey." and "Know who published the skills, agents, and CLAUDE.md / AGENTS.md files your AI runs, and verify that neither they nor the scripts they ship have changed since they were signed."

PromptSign provides origin authentication, integrity, accountability, and revocability. It takes a SHA-256 fingerprint of every file in a skill, including scripts, and signs those fingerprints under a verified identity from GitHub, Google, Microsoft, or a company account. The signature travels in a .promptsign/bundle.json file. The site says: "PromptSign takes a tamper-evident fingerprint (a SHA-256 hash) of every file in the skill, including any scripts it ships." and "The signature travels with the skill in a small .promptsign/bundle.json file. To check it, PromptSign re-fingerprints every file and confirms the signature, no server or internet connection required."

It offers a command-line tool and in-browser tools. The CLI plugs into Claude Code & Codex hooks to block unsigned or tampered instructions. The in-browser tools perform the same signing and verification, and file content does not leave the user's computer. PromptSign states that a signature proves who published a file and that it has not changed, but not that it is safe: "A signature proves who published this and that it hasn't changed. It does not prove that it's safe."