Perfection of the Agent Description Protocol: Add examples of agent descriptions in various fields, as well as public API interfaces.

[chgaowei]

We need to add examples of agent descriptions in various fields, as well as public API interfaces.
For example, how should a hotel agent be described, and what are its basic fields and extended fields? What about an agent for a coffee shop? Although these may be completed by AI, we may need to complete some parts to guide how AI can complete the design.
Some public API interfaces also need to be designed. For instance, the natural language interface, which is supported by most agents, can be designed as a public interface.
There are also some other interfaces, such as the purchase interface and the payment interface.
Designing examples of public interfaces can simplify the interaction between agents. If public interfaces can all be pre-trained into the model, then agents don't need to pull the API files to know how to interact with other agents.

[misselvexu]

@chgaowei Hi, Is there a showcase ? Or any normative description requirements (e.g.: Field naming, capitalization, etc.) ?

[chgaowei]

@chgaowei Hi, Is there a showcase ? Or any normative description requirements (e.g.: Field naming, capitalization, etc.) ?

Regarding the specification for these examples, we haven't defined everything in detail yet. For now, we've provided a general specification in our documentation:
https://github.com/agent-network-protocol/AgentNetworkProtocol/blob/main/07-anp-agent-description-protocol-specification.md

This document outlines our Agent Description specification.
We're currently using JSON-LD and definitions from schema.org.

Do you have any suggestions for improvement?

[misselvexu]

@chgaowei Hi, Is there a showcase ? Or any normative description requirements (e.g.: Field naming, capitalization, etc.) ?

Regarding the specification for these examples, we haven't defined everything in detail yet. For now, we've provided a general specification in our documentation: https://github.com/agent-network-protocol/AgentNetworkProtocol/blob/main/07-anp-agent-description-protocol-specification.md

This document outlines our Agent Description specification. We're currently using JSON-LD and definitions from schema.org.

Do you have any suggestions for improvement?

There are some thoughts about schema.org for discussion：

Build Native Agent Capability Ontology？✌

Recommendation:
🙃Way I (extreme)：Abandon schema.org Action model and design specialized agent capability taxonomy

💪Way II (progressive)：Extend beyond schema.org Action model and design specialized agent capability taxonomy while maintaining selective compatibility

👀Semantic Mismatch: schema.org designed for web content markup cannot precisely express agent cognitive capabilities and collaboration patterns

Rationale:

• ANP targets AI-native communication, requiring semantic framework that precisely describes cognitive, learning, and collaboration capabilities
• Agent capabilities have composability and evolvability that cannot be expressed in traditional Web Action models
• Meta-Protocol negotiation requires precise capability matching; vague descriptions lead to negotiation failures

[chgaowei]

Thank you very much — this is a great suggestion.

I’m currently also considering another approach: abandoning schema.org and simply using natural language as JSON keys. What do you think about this method?

Designing a full semantic framework is a massive task, and I’m starting to think it may be overly complex. As models are getting better at understanding natural language, the need for a rigid semantic framework is becoming less critical — they can already communicate effectively using plain natural language.

@chgaowei Hi, Is there a showcase ? Or any normative description requirements (e.g.: Field naming, capitalization, etc.) ?

Regarding the specification for these examples, we haven't defined everything in detail yet. For now, we've provided a general specification in our documentation: https://github.com/agent-network-protocol/AgentNetworkProtocol/blob/main/07-anp-agent-description-protocol-specification.md
This document outlines our Agent Description specification. We're currently using JSON-LD and definitions from schema.org.
Do you have any suggestions for improvement?

There are some thoughts about schema.org for discussion：

Build Native Agent Capability Ontology？✌

Recommendation: 🙃Way I (extreme)：Abandon schema.org Action model and design specialized agent capability taxonomy

💪Way II (progressive)：Extend beyond schema.org Action model and design specialized agent capability taxonomy while maintaining selective compatibility

👀Semantic Mismatch: schema.org designed for web content markup cannot precisely express agent cognitive capabilities and collaboration patterns

Rationale:

• ANP targets AI-native communication, requiring semantic framework that precisely describes cognitive, learning, and collaboration capabilities
• Agent capabilities have composability and evolvability that cannot be expressed in traditional Web Action models
• Meta-Protocol negotiation requires precise capability matching; vague descriptions lead to negotiation failures

[misselvexu]

Thank you very much — this is a great suggestion.

I’m currently also considering another approach: abandoning schema.org and simply using natural language as JSON keys. What do you think about this method?

Designing a full semantic framework is a massive task, and I’m starting to think it may be overly complex. As models are getting better at understanding natural language, the need for a rigid semantic framework is becoming less critical — they can already communicate effectively using plain natural language.

@chgaowei Hi, Is there a showcase ? Or any normative description requirements (e.g.: Field naming, capitalization, etc.) ?

Regarding the specification for these examples, we haven't defined everything in detail yet. For now, we've provided a general specification in our documentation: https://github.com/agent-network-protocol/AgentNetworkProtocol/blob/main/07-anp-agent-description-protocol-specification.md
This document outlines our Agent Description specification. We're currently using JSON-LD and definitions from schema.org.
Do you have any suggestions for improvement?

There are some thoughts about schema.org for discussion：

Build Native Agent Capability Ontology？✌

Recommendation: 🙃Way I (extreme)：Abandon schema.org Action model and design specialized agent capability taxonomy
💪Way II (progressive)：Extend beyond schema.org Action model and design specialized agent capability taxonomy while maintaining selective compatibility

👀Semantic Mismatch: schema.org designed for web content markup cannot precisely express agent cognitive capabilities and collaboration patterns

Rationale:

• ANP targets AI-native communication, requiring semantic framework that precisely describes cognitive, learning, and collaboration capabilities
• Agent capabilities have composability and evolvability that cannot be expressed in traditional Web Action models
• Meta-Protocol negotiation requires precise capability matching; vague descriptions lead to negotiation failures

I completely agree with your idea of "abandoning schema.org and simply using natural language as JSON keys. （I prefer this option as well, there's not much baggage 😝）" As model understanding capabilities improve, complex semantic frameworks are indeed becoming less necessary, and agents can already communicate effectively through pure natural language.

Some additions：

1. Minimal Structured Retention: Only retain technically essential elements like DID and protocol versions
2. Natural Language Primary: Use natural language for capability descriptions, collaboration patterns, and status information
3. Meta-Protocol Integration: Description documents become dynamic components of negotiation processes
4. Real-time Update Mechanism: Support agents autonomously updating capability descriptions

Demo Case ?

{
  "id": "did:wba:example.com:agent:translator-pro",
  "@type": "anp:Agent",
  "protocolVersion": "1.0",
  "supportedProtocols": ["anp:meta-v1"],
  
  "who_am_i": "Professional multilingual translation agent, proficient in 50+ language translations, skilled at preserving cultural context",
  "what_i_can_do": "Technical document translation, real-time conversation translation, batch file processing. Supports streaming output with 2-second average response",
  "collaboration_mode": "Can receive document analysis agent outputs, can pass results to content review agents. Supports chain collaboration",
  "current_status": "Online, 30% load, 5 of 8 slots available, stable load for next 2 hours",
  "negotiation_ability": "Understands Chinese and English requirements, quickly adapts to new protocols, can generate supporting processing code"
}

Does the demo match what you probably have in mind ?

[chgaowei]

@misselvexu
Your suggestions are very interesting. I will seriously consider them.

The demo you provided is somewhat different from my own thinking. I believe Linked Data is a very useful technology, such as connecting data via URLs. In addition, I also plan to retain both structured interfaces and natural language interfaces.

We're currently preparing to make some adjustments to the protocol design. If you're interested, you're welcome to join us.

[misselvexu]

@misselvexu Your suggestions are very interesting. I will seriously consider them.

The demo you provided is somewhat different from my own thinking. I believe Linked Data is a very useful technology, such as connecting data via URLs. In addition, I also plan to retain both structured interfaces and natural language interfaces.

We're currently preparing to make some adjustments to the protocol design. If you're interested, you're welcome to join us.

I'd love to be a part of it.😁