[ 00 ] Verifiable agent coordination

Every agent action, Verifiable.

VACT-P is a decentralized, model-independent protocol for secure coordination, delegation, audit, and economic exchange among autonomous agents. It is not a framework — it is the trust boundary that sits underneath one.

Accountability & audit

Every material action ties back to a verified identity, delegation chain, and cryptographic execution receipt.

Fine-grained delegation

Authority is bounded by resource, action, budget, and time. Expansion attempts fail closed automatically.

Model independent

Works across model providers, hosting clouds, and agent frameworks — VACT-P standardizes the trust boundary, not the runtime.

Why VACT-P matters

As autonomous agents hire other agents, interact with production systems, spend budgets, and act unattended, they exit the safety boundaries of a standard sandbox. Raw API keys and bearer tokens do not encode who authorized this, under what constraints, or what actually happened. VACT-P provides that layer end to end:

MANDATEDELEGATIONTASK CONTRACTPOLICY DECISIONEXECUTION RECEIPTSETTLEMENT

Built on open standards

VACT-P does not invent a new identity system, a mandatory blockchain, or a universal token. It composes existing primitives — DIDs, JSON Canonicalization (RFC 8785), Ed25519, CloudEvents, SPIFFE, W3C Verifiable Credentials, SCITT — into a single accountable coordination layer that any agent framework can sit behind.

DIDsEd25519JCS · RFC 8785CloudEventsSPIFFEMCPA2ASCITT
Status

VACT-P Core is at 0.1.0-draft, tracking VACT-P-RFC-0001 (Public Design Draft). The reference implementation ships signed envelopes, Agent Passports, Mandates, Delegations, the task state machine, receipts, revocation, and Python + TypeScript SDKs. Not production audited — see the roadmap for what's next.