Acknowledgements
agentrc is deliberately a thin declaration, packaging, and governance layer over proven open standards rather than a reinvention of them. The Agentfile is Dockerfile-shaped — four new keywords (IDENTITY, CAPABILITY, SOP, POLICY) over standard Dockerfile keywords — and it compiles to an OCI artifact carrying ai.agentrc.* labels. It owes a direct debt to the projects below — listed, where possible, by their source repositories. agentrc declares and governs these; it does not replace any of them.
Standards agentrc builds on
| Project | What agentrc adopts from it | Source |
|---|---|---|
| Agent SOP (Strands Agents) | An influence on the SOP keyword: a natural-language, RFC-2119-constrained operating procedure authored in the Agentfile and embedded as a readable file at /mnt/SOP (the label is a pointer + digest, never the full text). |
github.com/strands-agents/agent-sop |
| microsandbox | A reference for the deferred secrets design (host-scoped credential resolution). Secrets are out of scope for this draft — no SECRET/CRED keyword and no label schema yet. |
github.com/superradcompany/microsandbox |
| Model Context Protocol (MCP) | The open protocol for model/tool context. MCP servers are projected under /mnt/mcp/; agentrc declares and governs MCP, it does not replace it. |
github.com/modelcontextprotocol · modelcontextprotocol.io |
| Cedar (AWS) | The platform-side enforcement engine and compilation target for typed POLICY requests (never an Agentfile author surface); the AgentRC:: / Cedar authorization vocabulary. The platform compiles granted requests plus its own org rules into Cedar and enforces deny-by-default, forbid over permit, and tightening-only FROM composition. |
github.com/cedar-policy/cedar · cedarpolicy.com |
| Agent Skills | The open SKILL.md skill-bundle format. Skill bundles are projected under /mnt/skills/. |
github.com/agentskills/agentskills · agentskills.io |
Composes with
agentrc references these standards for packaging, signing, provenance, observability, and multi-agent orchestration. They are not required by the core, but conforming platforms and registries are expected to lean on them.
| Project | Role in agentrc | Source |
|---|---|---|
| Open Container Initiative (OCI) | The artifact format: standard layers carry the /mnt resources and the image config carries the ai.agentrc.* labels; application/vnd.agentrc.* media types. |
github.com/opencontainers · opencontainers.org |
| Sigstore | Package signing and verification. | github.com/sigstore · sigstore.dev |
| SLSA | Supply-chain provenance for packages. | github.com/slsa-framework · slsa.dev |
| OpenTelemetry | Audit and telemetry sinks reached via POLICY agent.hooks.* and agent.telemetry_sink requests. |
github.com/open-telemetry · opentelemetry.io |
| A2A (Agent2Agent) | A reference point for the deferred multi-agent workflow companion. The agent-to-agent protocol is out of scope for this version; capability exposure via IDENTITY / CAPABILITY / labels is in scope. |
github.com/a2aproject/A2A |
Naming
agentrc is the specification. AIO is the reference implementation and test harness. The relationship is the same as ECMAScript to V8, or OCI to runc — the standard is separate from any one implementation of it.