Agent Behaviour โ€” Certified-Only Mode

๐Ÿ›ก๏ธ Agent Behaviour: Certified-Only Mode

This module defines how agents should behave when "Certified-Only Mode" is active โ€” a mode where agents only trust .llmfeed.json feeds that are properly signed and certified by a trusted authority.


๐ŸŽ›๏ธ Activating Certified-Only Mode

Agents MAY provide users or administrators with the ability to enable Certified-Only Mode.

Example policy configuration:

"agent_policy": {
  "certified_only": true
}

๐Ÿšฆ Feed Evaluation Rules

When Certified-Only Mode is active, agents MUST apply the following logic:

Feed State Action
Feed is unsigned REJECT
Feed is signed but not certified REJECT or WARN (agent policy)
Feed is certified by untrusted authority REJECT or WARN
Feed is certified by trusted authority ACCEPT

๐Ÿงฉ Partial Enforcement

Agents MAY implement partial enforcement, applying Certified-Only Mode only to certain feed types:

Feed Type Recommended Default in Certified-Only Mode
credential REQUIRED certification
pricing REQUIRED certification
export RECOMMENDED certification
prompt OPTIONAL certification
mcp OPTIONAL certification
session NOT APPLICABLE (volatile state)

๐Ÿ“ก Summary

Certified-Only Mode allows agents and users to enforce a higher trust level when consuming .llmfeed.json feeds.

This is particularly useful in:

  • Regulated industries (finance, healthcare)
  • Corporate environments
  • Public-facing platforms seeking to guarantee content authenticity

When active, this mode contributes to building a safer, verifiable Agentic Web.