This package is a pointer, not the CLI.
PromptSign signs and verifies AI instruction files — skills, agent definitions,
CLAUDE.md, AGENTS.md — and the script payloads that ship with them.
The CLI is a single static binary, distributed as a signed release artifact rather than through a package manager. It runs inside agent hooks on every session start and skill invocation, so it has to start in milliseconds and require no runtime on the machine.
Download it from the project homepage: https://promptsign.ai/docs
The verifier is a trust root for everything that relies on it. Installing one through a channel that executes install scripts, from a package that could be replaced, is the wrong shape for that job — which is the same argument PromptSign makes about instruction files.
To verify in-process from Node — for a marketplace, registry, or CI check — use the native binding:
npm install @promptsign/verifyAgentic coding tools load instruction files directly into a model's context, where they act as executable code, and skills bundle scripts the host actually runs. They are distributed through unvetted channels with no origin authentication, no integrity protection, and no revocation.
PromptSign is the wire format that supplies all three, plus the verifiers that enforce it. The specification is published separately, so implementations are replaceable and the format is the thing others standardize on: https://github.com/PromptSign/spec
A valid signature establishes origin and integrity — not safety. A signed prompt-injection payload is still a prompt-injection payload.
Apache License 2.0.