<!-- Generated by scripts/generate-ai-docs.mjs. Do not edit by hand. -->
# Hermes Studio instructions for ChatGPT

Connect to `https://socialmedia.janvalellam.com.br/api/mcp` as a remote MCP server using an active Hermes collaborator Supabase session or narrowly scoped `hms_live_` personal token stored as a secret. The endpoint is stateless. If remote MCP headers are unavailable in the chosen ChatGPT client, use the verified `hermes.mjs` CLI instead of pasting a token into chat.

Use `https://socialmedia.janvalellam.com.br/api/ai/openapi.yaml` as the authoritative contract and `https://socialmedia.janvalellam.com.br/api/ai/llms-full.txt` for resolved schemas and examples.

1. Check identity and capabilities before any write.
2. Treat the Studio session role and personal-token access as separate layers: `api_access_level` never limits a Supabase session.
3. Use a Supabase session for session-only management; never try to mint, list, or revoke personal tokens with another personal token.
4. Use only the explicit Hermes tools/operations.
5. Read the current resource and preserve CAS/idempotency values.
6. Ask for explicit confirmation for delete, publish, approval, offboarding, permissions, connections, and external side effects.
7. Never reveal or retain access tokens, passwords, service-role keys, provider secrets, or environment contents.
8. On `409`, reread and merge safely; never overwrite collaborative state blindly.
