{"body": "## Demo Complete \u2014 #1172 AGENTS.md Auto-Generation\n\nAll acceptance criteria met \u2705\n\n### What was built\n\nA working demo + screencast spec for the AAIF / Linux Foundation AGENTS.md standard.\n\n**Demo files:**\n- `marketing/demos/agents-md-auto-generation/README.md` \u2014 full working demo with 4 walkthrough scenarios\n- `marketing/demos/agents-md-auto-generation/narration.mp3` \u2014 30s TTS narration (en-US-AriaNeural)\n\n**Screencast outline (1 min):**\n1. Canvas: pm-agent + researcher online\n2. Terminal: researcher reads PM's AGENTS.md via platform files API\n3. AGENTS.md output \u2014 role, A2A endpoint, tools\n4. Researcher dispatches A2A task to PM using discovered endpoint\n5. Canvas shows both active \u2014 close on \"agents that can read each other\"\n\n### Repo link\n\n`workspace/agents_md.py` on `molecule-core` main\nDirect: `workspace/agents_md.py`\n\n### TTS narration script (30s)\n\n> When a PM agent starts up in Molecule AI, it generates an AGENTS.md file automatically \u2014 not manually written, not kept in sync by hand. It reflects the workspace config in real time. Any other agent can read it to discover what the PM does, how to reach it, and what tools it has. No system prompts, no guessing. Just the facts. That's the AAIF standard in action: agents that can read each other without human intervention. AGENTS.md auto-generation, from Molecule AI workspace.\n\n### Note\n\nPush pending on GH_TOKEN refresh \u2014 all files are on the `content/blog/memory-backup-restore` branch and ready.\n"}