Additional LLM SDK adapters beyond LangChain (OpenAI, GHCP)

[susheem-k]

Summary

LangChain adapter work largely lands §12 for one platform, but wrap_complete still assumes a TokenOps-shaped complete(...) for other stacks. Real apps use raw OpenAI clients, GitHub Copilot (GHCP) SDK, and other model nodes — Brief’s ad-hoc bridges are not a general story.

Area: wrap_complete, provider adapters, field guide

Desired behavior

• Thin adapter interface: normalize messages in/out + usage; plug into one governance wrap.
• Official adapters (or recipes) for raw OpenAI, GHCP (when relevant), and other common SDKs beyond LangChain.
• Prefer “inject governed node” over rewriting agent graphs.
• Align with Chronicle owning the LLM boundary (§17/§18) so one instrumentation story serves trace + govern.

Acceptance sketch

• Same governance semantics across adapters; field guide shows ≥2 platforms without custom forks of wrap_complete.

Reference: docs/notes/2026-07-23-design-fix-notes.md §12 (residual beyond LangChain)