🧩 LLMFeed Extensions

This document lists optional fields, experimental structures, and advanced practices that extend the core LLMFeed protocol.

They are not mandatory — but agents may support and react to them.

✅ Core extensions

⚠️ `flags[]`

Flags can be added by agents or moderators to warn about:

• mismatched capabilities
• deprecated versions
• suspicious or malicious usage

"flags": [
  {
    "type": "risk",
    "submitted_by": "agent://previewbot",
    "reason": "Declared actions not matching real API",
    "date": "2025-05-19T12:00:00Z",
    "status": "pending",
    "source": "https://llmca.org/flag/234"
  }
]

🔐 Trust extensions

Agents may use:

• trust.scope = certified
• certifier = https://llmca.org
• signed_blocks = [ "metadata", "prompts", "trust" ]

Verification interfaces (like /verify) will display results clearly.

🧠 Agent behavior hints

Declare how the LLM should behave:

"agent_behavior": {
  "mode": "suggest-only",
  "requires_user_confirmation": true
}

🧪 Runtime policies (experimental)

You can expose:

"sandbox": {
  "max_calls": 10,
  "device_scope": "local-only",
  "runtime_constraints": "No background tasks"
}

Or create a separate sandbox-policy.llmfeed.json.

📚 Related

• `llmfeed.md`
• `agent-behaviour`
• `feedtype_manifesto.md`
• `feedtype_capabilities.md`
• `feedtype_llm-index.md`