T045: iris-adapter-llm-openai

[likith1908]

Summary

Implements T045 - the iris-adapter-llm-openai workspace package providing the OpenAI LLM adapter. Extends OpenAICompatProvider from T043 (iris_adapter_llm_shared). Stacked on #43.

• packages/iris-adapters/llm-openai/src/iris_adapter_llm_openai/client.py: OpenAIProvider implementing id, _base_url(), _auth_headers(), and _model_for_hint(). Single endpoint https://api.openai.com/v1/chat/completions. Auth via Authorization: Bearer header. Two model slots (chat + extract) selected by model_hint field, no ContextVar needed (model goes in request body, not URL).
• packages/iris-adapters/llm-openai/tests/test_unit.py: 23 unit tests covering all C-LLM-001..010 contract points using a mock httpx.AsyncClient.
• packages/iris-adapters/llm-openai/tests/test_live.py: 3 live tests gated on IRIS_LLM_LIVE_OPENAI=1, verified against real OpenAI endpoint.
• pyproject.toml: iris_adapter_llm_openai registered in importlinter root_packages, layers, independence, forbidden, mypy strict overrides, and coverage source.
• tests/001-project-scaffold/: workspace, import-linter, and mypy scaffold tests updated for the renamed module.
• .env.example: IRIS_LLM_OPENAI_API_KEY and optional model env vars documented.

Task reference

• Task ID: T045
• Workstream: 004-llm-adapter-set

Acceptance criteria

T045

• OpenAIProvider.from_env() constructs from env vars; raises RuntimeError if IRIS_LLM_OPENAI_API_KEY is missing
• Endpoint: https://api.openai.com/v1/chat/completions
• Auth: Authorization: Bearer <key> header, no api-key header
• model_hint="extraction" routes to model_extract; all other hints route to model_chat
• All C-LLM-001..010 contract points covered by unit tests
• Live tests pass against real OpenAI (IRIS_LLM_LIVE_OPENAI=1): 3/3 passed

PR checklist

• Every acceptance criterion in the task's tasks.md entry is satisfied.
• docs-ci workflow passes (markdown lint + tasks structural check).
• No em-dashes introduced in new prose.
• All internal links resolve.
• No secrets, credentials, or personal names introduced.

Rebase note

This PR is stacked on #43. Once #43 merges, rebase onto main before merging this one.

Notes for the reviewer

Module rename: Scaffold was iris_llm_openai; renamed to iris_adapter_llm_openai to match the iris_adapter_* naming convention.

No ContextVar: Unlike Azure, OpenAI sends the model in the request body JSON field, so no per-request routing state is needed. _model_for_hint() simply returns the model string and _base_url() always returns the same constant URL.

Live tests: Verified against real OpenAI endpoint with gpt-4o-mini. All 3 tests passed.

[anmolg1997]

Reviewed on the stack (own diff against T044-llm-azure-openai). Same verdict as the Azure adapter: a clean minimal subclass that proves out the base class.

• Constant endpoint https://api.openai.com/v1/chat/completions and Authorization: Bearer header. Correct.
• _model_for_hint routes extraction vs chat to the two configured models; everything else inherited from the tested base.
• from_env / _require_env mirror the Azure adapter, no credential leak.

Coverage 94% with the single miss being the _require_env raise, consistent with the others. Error mapping, retry, and structured output are all inherited from llm-shared and already covered there, so there is nothing adapter-specific left untested here.

Stacked on #43, so it rebases up the chain as the zipper proceeds. Approving the content; will re-confirm on rebase onto main.