For AI agents

Every SocialClaw surface an AI agent can consume — llms.txt, llms-full.txt, OpenAPI, the MCP server, and the JSON-emitting CLI — plus the five conventions well-behaved agents follow when publishing social media.

Every surface a model might consume, generated from the live contract:

• llms.txt and llms-full.txt — plain-text product and API descriptions written for language models.
• openapi.json — OpenAPI 3.1 with JSON Schema 2020-12 components.
• MCP server — 31 workspace-scoped tools, hosted and stdio.
• CLI — every command emits machine-readable JSON with --json.

What well-behaved agents do:

1. Validate before apply. validate_schedule / POST /v1/posts/validate creates nothing and names problems per post.
2. Always send an idempotencyKey. Agents retry; the key makes retries safe instead of duplicating runs.
3. Discover, don't assume. account_capabilities reports each account's real media and text limits.
4. Verify outcomes. A created run is not a published post — check run_status after the publish time, and read post_attempts on failures.
5. Reuse hosted media. list_assets surfaces files the user already uploaded — prefer them over re-uploading.