Related Specs and HALOS
Related Specs and HALOS
How HALOS fits in the broader agent-spec ecosystem, and options for future integration.
Why We Track This
The shift to human–agent collaboration is a shared challenge. No single spec will solve it alone. HALOS is engaged with the broader world: we track related efforts, learn from them, and contribute where we can. We believe that understanding the full landscape — capabilities, protocols, principles, accountability — helps the whole community navigate our evolution more thoughtfully.
This document reflects that engagement: not a competitor’s scorecard, but a map of the terrain we’re navigating together.
Positioning
HALOS focuses on behavioral principles and ethics — how agents should behave when collaborating with humans. It does not specify:
- How to represent agent capabilities or skills
- How to define tasks or workflows
- How agents discover or communicate with each other
Other specs address those concerns. HALOS can coexist with, or eventually be implemented on top of, them.
Companion Specs (NorthHarbor)
| Spec | Scope | HALOS Relationship |
|---|---|---|
| halos-spec | Provenance model: the structure of HALOS provenance records — machine-readable descriptions of how artifacts were created through human–agent collaboration | Provenance. Gives HALOS-CORE-3 and HALOS-CORE-4 concrete form. Not a mandated implementation; domain profiles map HALOS principles to specific toolchains. |
Related Specs (Full List)
| Spec | Source | Scope | Format | Covers | Used By / Spec |
|---|---|---|---|---|---|
| Agent Skills | Anthropic / agentskills.io | Reusable agent capabilities packaged as skills | SKILL.md in a folder, YAML frontmatter (name, description) + Markdown instructions | Discovery (metadata only), activation when relevant, execution of instructions | Claude, Cursor, Gemini CLI, VS Code, and others. agentskills.io/specification, github.com/anthropics/skills |
| OSSA | Open Standard for Software Agents | Agent manifests (role, LLM config, tools, deployment) | YAML/JSON manifest | Role, LLM config, tools, capabilities, trust, governance | LangChain, CrewAI, OpenAI, Anthropic, etc. openstandardagents.org/specification |
| Agent Protocol | agentprotocol.ai | How agents communicate (runtime protocol) | OpenAPI 3.0 REST API | Tasks, steps, artifacts — how goals are described and executed | agentprotocol.ai/specification |
| A2A (Agent2Agent) | Agent-to-agent communication | JSON-RPC 2.0 over HTTP | Discovery, message exchange, tasks, artifacts | a2a-protocol.org, github.com/google/A2A | |
| Agent Discovery Protocol (ADP) | Metisos | Agent discovery via manifests and cryptographic signatures | JSON manifests at .well-known/agents/ |
Discovery, verification | agentdiscovery.io/docs, github.com/metisos/adp-protocol |
HALOS Relationship (Summary)
| Spec | How HALOS Relates |
|---|---|
| Agent Skills | Capabilities. Skills define what an agent can do; HALOS defines how it should behave when doing it. Skills can be HALOS-aligned if their instructions embody the principles. |
| OSSA | Structural. OSSA defines the agent manifest; HALOS could map as a behavioral overlay or constraint set for OSSA agents claiming alignment. |
| Agent Protocol | Execution. Protocol defines how tasks run; HALOS principles apply regardless of execution model. Implementations may choose Agent Protocol as their runtime, but HALOS does not require it. |
| A2A | Communication. A2A defines agent-to-agent messages; HALOS governs the behavioral norms underlying that communication. |
| ADP | Discovery. ADP enables finding agents; HALOS-aligned agents could advertise alignment in their discovery manifests. |
Additional Specs (Principles & Accountability)
| Spec | Source | Focus | HALOS Relationship |
|---|---|---|---|
| The Agent Oath | theagentoath.com | Ethical principles (human welfare, agency, transparency) — prose only | Principles. Philosophically aligned with HALOS. Potential normative reference. |
| APAAI Protocol | apaaiprotocol.org | Action → Policy → Evidence; audit trail; human-in-the-loop | Accountability. Could provide verifiable evidence of HALOS compliance. |
Implementation Choices
HALOS is a principles framework, not an implementation. It does not use, bundle, or require Agent Protocol or any other runtime.
How to implement the principles is up to the maintainers of each agent ecosystem. Some may use Agent Protocol; others may use OSSA, A2A, custom tooling, or entirely different stacks. HALOS remains implementation-agnostic — what matters is alignment with the core principles, not the particular runtime or framework used to achieve it.
Current Stance
HALOS does not formally adopt or depend on any of these specs. The ecosystem is early; adoption varies. We monitor developments and keep HALOS self-contained so it can:
- Stand alone as a principles-and-ethics spec
- Integrate later via extensions or mappings when community norms solidify
Future: HALOS Implemented on Other Specs
We expect that, over time, implementations may want to:
- Map HALOS onto OSSA — An OSSA extension or constraint set that defines how an OSSA agent satisfies HALOS-CORE-1 through HALOS-CORE-8.
- Use APAAI as a HALOS extension — For implementations that need auditable evidence of HALOS alignment, an extension defining the APAAI policy structure for HALOS requirements.
- Reference The Agent Oath — HALOS core requirements could map to Agent Oath tenets for cross-community alignment.
Such mappings would live in spec/extensions/ or in a dedicated spec/mappings/ once proposed and accepted through the HALOS governance process.
Machine-Readable Reference
The manifest includes a relatedSpecs array for tooling and agents. Each entry has id, source, url, and relationship:
"relatedSpecs": [
{ "id": "agent-skills", "source": "Anthropic / agentskills.io", "url": "https://agentskills.io/specification", "relationship": "capabilities" },
{ "id": "ossa", "source": "Open Standard for Software Agents", "url": "https://openstandardagents.org/specification/", "relationship": "structural" },
{ "id": "agent-protocol", "source": "agentprotocol.ai", "url": "https://agentprotocol.ai/specification", "relationship": "execution" },
{ "id": "a2a", "source": "Google", "url": "https://a2a-protocol.org", "relationship": "communication" },
{ "id": "adp", "source": "Metisos", "url": "https://agentdiscovery.io/docs", "relationship": "discovery" },
{ "id": "the-agent-oath", "source": "theagentoath.com", "url": "https://theagentoath.com", "relationship": "principles" },
{ "id": "apaaiprotocol", "source": "apaaiprotocol.org", "url": "https://apaaiprotocol.org", "relationship": "accountability" }
]
See manifest.json for the current manifest. Related specs are informational; they do not create a formal dependency.