Text
Width

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:

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.

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) Google 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:


Future: HALOS Implemented on Other Specs

We expect that, over time, implementations may want to:

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.