Add a dedicated ChatGPT gateway runtime

[yia-mw-agent]

Parent

• PRD: Dedicated FAVA Data Repo MCP Tunnel for ChatGPT #67

What to build

Add an explicit long-lived FAVA Trails gateway runtime for ChatGPT access through OpenAI Secure MCP Tunnel. The runtime is owned by a single FAVA data repo, not by an agent session or product working directory. Existing fava-trails-server stdio behavior must remain the default for Claude, Codex, Hermes, and other per-agent clients.

The gateway should require an explicit FAVA_TRAILS_DATA_REPO, initialize against that repo, validate startup prerequisites, and produce operator-readable startup output. It should not introduce a public HTTPS endpoint, per-machine process registry, dynamic agent registration, or MachineWisdom-specific host assumptions.

Acceptance criteria

• A dedicated gateway command or run profile starts one long-lived MCP runtime for one configured FAVA data repo.
• Default fava-trails-server behavior remains unchanged for existing stdio MCP clients.
• Gateway startup refuses or clearly fails when JJ is unavailable, FAVA_TRAILS_DATA_REPO is missing, the data repo is unreachable, or required Trust Gate provider config is missing.
• Gateway startup output identifies the data repo, trails directory or equivalent runtime scope context, and the fact that it is intended for OpenAI Secure MCP Tunnel usage.
• The gateway does not infer scope from a caller working directory; scoped tools still require explicit trail_name.
• The implementation is deployment-neutral and contains no WiseMachine hostnames or MachineWisdom-only infrastructure assumptions.

Blocked by

None - can start immediately