Studio agent sessions (chat / book-create) don't send `x-opencode-session`, so OpenCode Go returns 400 MissingSessionID

Studio agent sessions (chat / book-create) don't send x-opencode-session, so OpenCode Go returns 400 MissingSessionID.

Env

InkOS 1.8.0 (latest, master same), Node v24, Windows. OpenCode Go as a custom service: baseUrl=https://opencode.ai/zen/go/v1, apiFormat=chat, API key set. "Test connection" passes.

Problem

Starting a chat / book-create session in Studio fails:

400 Error from provider (Console Go): Request is missing x-opencode-session
and cannot be routed efficiently.

The pipeline (inkos write next) works. OpenCode Go has required x-opencode-session on every inference request since 2026-09-03.

Root cause

The pipeline path createLLMClient() (packages/core/src/llm/provider.ts) reads config.headers ?? parseEnvHeaders() and puts them on the pi-ai model, so llm.headers / INKOS_LLM_HEADERS work there.

The Studio agent path builds its model via resolveServiceModel() (packages/core/src/llm/service-resolver.ts), which never sets headers. That model is used by runAgentSession → guardedPiStream, which only forwards options.headers. So configured headers are silently dropped on the agent path.

pi-ai's openai-completions does merge model-level headers (const headers = { ...model.headers }), so setting model.headers is sufficient.

Proof

Direct request to the endpoint: without header → 400 MissingSessionID; with x-opencode-session → 200 OK.

Suggested fix

• service-resolver.ts: add optional extraHeaders?: Record<string, string> to resolveServiceModel and include ...(extraHeaders ? { headers: extraHeaders } : {}) in the built model.
• packages/studio/src/api/server.ts: pass config.llm.headers to the 3 resolveServiceModel(...) calls.

(CLI/TUI agent paths already use client._piModel, which has headers, so only Studio needs this.)

Workaround

A local proxy in front of baseUrl that injects x-opencode-session (verified working).