📜 CHANGELOG — LLMFeed / MCP Specification

[1.0.0] – 2025-05-21

📦 Finalization of LLMFeed Specification

  • Restructured the entire specification into a logical, numbered folder hierarchy
  • Added llmfeed.md as the canonical entry point for LLMFeed structure and usage
  • Introduced formal specs for new feed types:
    • session.llmfeed.json
    • prompt.llmfeed.json
    • credential.llmfeed.json
    • pricing.llmfeed.json
  • Added wellknown.md to formalize feed discovery and .well-known/ structure
  • Clarified usage of audience, intent, session_state and agent_services
  • Updated block-reference.md with signature and certification blocks
  • Made canonicalization and trust logic more consistent with signed_blocks
  • Separated normative logic from site-specific features and ecosystem demos
  • Prepared for indexing, flagging, and signature verification via llmca.org
  • Replaced old V0.9 with a coherent, agent-ready, open standard foundation

This marks the first full public release of the Model Context Protocol (MCP) as a standard for structured, verifiable agent/web interoperability.

🧠 Agent Behavior & Prompt Injection (Behavioral Layer Finalization)

  • Introduced prompt-index.llmfeed.json and /.well-known/prompts/ usage as injectable capsules
  • Defined and published 4 official prompts for agent override and session export:
    • mcp-mode-activation.llmfeed.json
    • generate-session-feed.llmfeed.json
    • mcp-agent-behavior-override.llmfeed.json
    • prompt-index.llmfeed.json
  • Updated llmfeed_feedtype_prompt.md to clarify persistent prompt logic
  • Added intent_router[] inside mcp.llmfeed.json to route user expressions to specific behaviors
  • Refactored agent-behaviour.md to reflect prompt-based behavior injection and trust models
  • Added guidance for generate-session-feed.llmfeed.json in session export page
  • Aligned all /tools/* pages with real prompt usage (prompt, verify, behavior, session)

This version stabilizes how agents learn, behave and persist session logic using verifiable prompt capsules.

[UPCOMING] – Proposals under consideration

These are not included in 1.0.0 but may be added in future minor releases.

  • agent-profile.llmfeed.json to declare agent capabilities and behavior
  • Agent store & flagging graph (integration with LLMCA reputation system)
  • bundle.llmfeed.json for file archive summarization
  • feedtype_app for mobile/app exposure (linked to store metadata)
  • prompt.llmfeed.json signature and ownership assertions
  • Persistent session memory guidelines and tooling
  • ✅ Support for fallback_certifier in trust
  • ✅ OpenAPI integration block in capabilities[]
  • ✅ New flag llm_simplified: true for basic agent use
  • ✅ mcp_metadata.openapi_hint field for OpenAPI discoverability
  • ✅ Proposal for full-signature default in mcp.llmfeed.json