Feed Type: `capabilities.llmfeed.json`

Purpose

This feed lists live, callable actions that agents may invoke on behalf of users.
These capabilities represent exposed functions, API routes, or service endpoints.

Agents use them to understand what they can do, and how to call it — without guessing.

Location

Recommended path:

.well-known/capabilities.llmfeed.json

Each capability MUST declare:

a name
a method (GET, POST, etc.)
a path
a description

Canonical Fields

{
  "name": "submitContactForm",
  "method": "POST",
  "path": "/api/contact",
  "description": "Allow a user to request support or callback",
  "input_schema": {
    "required": ["name", "email", "message"]
  }
}

🧩 Optional fields

Field	Description
`input_schema`	Required fields or JSON schema
`llm_trust_level_required`	e.g. `"any-agent"`, `"certified-only"`
`rate_limit`	e.g. `5/min`
`requires_user_consent`	`true` or `false`
`audience`	e.g. `["llm"]`, `["developer"]`
`tags`	Classifier keywords

🧠 Agent behaviour

Agents must:

Only call capabilities they understand and are permitted to
Confirm if requires_user_consent = true
Validate input_schema before submitting
Adjust prompts or voice output based on description

✅ Optional Capabilities (Suggested)

These are not mandatory, but recommended to boost interoperability and real-world usefulness.

🤝 Support & Interaction

Name	Path	Description
`requestCallback`	`/callback`	Request a human to call back the user
`submitContactForm`	`/contact`	Submit a message or inquiry
`bookAppointment`	`/book`	Book a meeting or service
`getSupportAvailability`	`/support-status`	Return human presence availability

🌍 Localisation & Language

Name	Path	Description
`getRegionsServed`	`/regions`	Return ISO regions supported
`getLanguagesSupported`	`/languages`	Return languages agents can use

📦 Logistics & Reservations

Name	Path	Description
`trackDelivery`	`/track`	Track a parcel or service ETA
`getETA`	`/eta`	Estimate arrival time
`cancelReservation`	`/cancel`	Cancel a user reservation

🔐 Consent & Compliance

Name	Path	Description
`downloadPersonalData`	`/gdpr/export`	User can access their data
`submitConsentForm`	`/consent`	Required consent prior to action
`getWrapperAuditTrail`	`/audit`	Show trace of agent actions (transparent AI)

📎 Related

🧠 OpenAPI Compatibility

While the capabilities.llmfeed.json feed type is designed to offer a simple, high-level view for agents and LLMs, it may optionally reference a complete OpenAPI file for technical documentation.

This allows agents to:

detect the service intent via capabilities[].intent
consult precise calling schemas via the OpenAPI spec

Example combination:

{
  "feed_type": "capabilities",
  "metadata": { "title": "France-Care Capabilities" },
  "capabilities": [
    {
      "type": "endpoint",
      "intent": "check subscriber status",
      "url": "https://api.france-care.fr/abonnement",
      "description": "Returns current user subscription level"
    },
    {
      "type": "openapi",
      "url": "https://france-care.fr/.well-known/openapi.json",
      "description": "Detailed OpenAPI specification for this backend"
    }
  ]
}

MCP and OpenAPI are complementary:

MCP provides context, trust, and intent.
OpenAPI provides detailed function contracts.

Feed Type: capabilities.llmfeed.json