Charon does not score vague risk. It detects concrete patterns in typed agent actions before anything launches.
Charon is not just a blocker. Every decision writes a receipt with the action, policy hash, verdict, execution status, and signature.
Install Charon locally, gate commands through policy, then inspect the receipt trail.
npx github:CharonAI-code/charon setupCreates charon.yml, generates an Ed25519 identity, and installs the local charon command.charon gate -- npm testRuns a shell command through policy first. Charon returns PASS, PAUSE, or DENY before launch.charon receipts latestShows the latest trusted receipt so the decision can be inspected and verified later.