ColomboAIDOCS
API ReferenceCookbook
Docs/MC-1 Security/Security verification and Trust evidence
PublishedGuidanceLAST REVIEWED · 18 NOV 2026

Security verification and Trust evidence

Publish scoped effect evidence without confusing an applied action with independently current assurance.

Runtime scope

A separate verifier

The verifier uses a distinct expiring security:verify key. It obtains a60second challenge bound to its identity, tenant, response and revision. The identity authority supplies a read-only agent proof delegated by the current registered owner to this verifier, with empty permissions and a maximum five-minute horizon.

The verifier makes an authenticated HTTP request at the existing registered-agent authorization gate. Invalid identities, stale owners or credential epochs, another control and an ordinary403 cannot establish the expected effect. The server records its own gate observation; a client-submitted outcome is prohibited.

One exact verification

Supply MC1_SECURITY_URL, MC1_SECURITY_VERIFIER_KEY and MC1_SECURITY_AGENT_PROOF through the approved secret store. The verifier does not receive the dispatcher key or mint a target proof. The script prints only scoped receipt metadata.

node scripts/security-verification-worker.mjs RESPONSE_ID REVISION

Atomic Trust publication

Source observations and successful gate verifications use the existing encrypted Trust evidence vault with tenant/project/target/evidence/control associated data. Publication changes the Trust version and revokes stale certificates in the same transaction.

The verification receipt records the exact gate, observed owner/epoch, response, control, verifier and observation time. Its assurance expires within30seconds. Historical evidence retrieval within source retention does not renew assurance, and this receipt cannot independently certify the full Security platform or an agent.

Read and investigate

Current owners/admins and exact expiring security:read keys may retrieve scoped Security content. Ordinary usage keys and source credentials cannot. Content reads are audited; authority, retained data, key configuration and integrity are checked again before delivery.

The Console shows an evidence timeline and a separate status for current access verification. Treat a stale, missing, failed or inconclusive check as unverified; do not infer process or network isolation.

End of guide
Was this page helpful?
Security verification and Trust evidence | MC-1 Docs