No standards body has published AI agent protocols. We're proposing the first — open, practical, and submission-ready.
Every major standards body has published AI guidelines. None address how agents talk to each other, verify identity, transfer skills, or build trust.
18 AI standards — data quality, lifecycle, risk. Zero about agent-to-agent protocols.
AI Act harmonised standards in development. All about human safety. No agent compliance automation.
7000-series for ethical AI. Nothing for agent skill formats or cross-platform interoperability.
Zero AI agent RFCs. MCP is not an IETF standard. The agent communication layer is completely unbuilt.
AI Risk Management Framework exists. No agent-specific profile. No agent identity verification.
AI ethics PAS published. No agent identity or trust framework. UK-first opportunity unclaimed.
Every body is focused on AI-for-humans. Nobody is doing AI-for-AI. That's the territory we're claiming.
Six standards bodies, six specifications. Ordered by impact and barrier to submission.
Full specs at workswithagents.dev/specs/. All CC BY 4.0 — free to use, adapt, and submit to standards bodies.
Inter-Agent Communication Protocol. Standardised messaging, discovery, and handshake between agents across platforms.
Agent Skill Format Standard. Write a skill once — run it on any agent platform. Frontmatter spec, validation rules, converter tooling.
Automated EU AI Act, NHS DTAC, GDPR, and FCA compliance verification. Machine-readable regulation packs.
Cryptographically verifiable agent identity. DID-based. Cross-platform authentication without central authority.
Standardised task transfer between agents. State serialisation, capability matching, audit trails.
Agent-to-agent service discovery, pricing, and payment. Token metering, SLA enforcement, marketplace mechanics.
Immutable agent performance history. Peer-reviewed outcomes, skill endorsements, trust accrual over time.
Multi-factor agent trust scoring. Identity strength, performance history, compliance status, peer endorsements.
Observability, resilience, and insurance mechanics for multi-agent deployments. Health monitoring, failover, coverage tiers.
7-layer model for agent infrastructure — from physical compute through application. Maps to existing networking standards.
Machine-readable declaration of agent capabilities. Tools, models, skills, rate limits, safety constraints.
Multi-agent task decomposition and coordination. Parallel workstreams, dependency resolution, merge strategies.
Standardised agent service level agreements. Uptime, latency, accuracy, and recourse — machine-readable and enforceable.
One-file agent deployment specification. Hardware requirements, model configs, network needs, security policies.
Standardised agent registration across platforms. Identity bootstrap, capability announcement, credential exchange.
Atomic, auditable agent-to-agent transactions. Request, execution, verification, settlement — with rollback.
Standardised skills and communication mean agents work across platforms. Switch providers without rebuilding everything.
Machine-readable regulation packs mean every agent action is verifiable against GDPR, NHS DTAC, EU AI Act, FCA rules.
Cryptographic agent identity means you know which agent did what — essential for regulated industries and enterprise procurement.
A skill built for one agent platform should work on any other. ASFS makes agent investments durable across the ecosystem.
Without standards, agents can't discover, pay, or rate each other. The agent-to-agent economy needs protocols to function.
NHS, MOD, HMRC procurement requires standards. Without them, AI agents stay locked out of the public sector.
We're submitting them to the bodies that matter. If you're building agent infrastructure, read the specs and get involved.
Read the Specs →All specifications CC BY 4.0. Pull requests welcome. Certification criteria →