From 2583842e21963099c65c709732677a716c03e7ff Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Sat, 13 Jun 2026 22:39:11 +0000
Subject: [PATCH 1/2] Type llm messages with ChatCompletionMessageParam instead
 of suppressing

The gateway `complete()` call carried a bare `# type: ignore[arg-type]`
because `build_messages` returned `list[dict[str, str]]`, which isn't
assignable to the OpenAI SDK's `Iterable[ChatCompletionMessageParam]`.
Type the messages list and both function boundaries with the SDK's own
param type so the `create()` call type-checks without the suppression.
Tighten one build_messages test to compare the whole dict (matching its
siblings), avoiding a not-required-key subscript on the now-typed union.
---
 aai_cli/core/llm.py | 10 +++++-----
 tests/test_llm.py   |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/aai_cli/core/llm.py b/aai_cli/core/llm.py
index 19b5747b..0c8cb0c7 100644
--- a/aai_cli/core/llm.py
+++ b/aai_cli/core/llm.py
@@ -9,7 +9,7 @@
 
 if TYPE_CHECKING:
     from openai import OpenAI
-    from openai.types.chat import ChatCompletion
+    from openai.types.chat import ChatCompletion, ChatCompletionMessageParam
 
 # The LLM Gateway is OpenAI-compatible, so we talk to it through the OpenAI SDK
 # pointed at the active environment's gateway base (see _client / code_gen).
@@ -76,7 +76,7 @@ def build_messages(
     system: str | None = None,
     transcript_id: str | None = None,
     transcript_text: str | None = None,
-) -> list[dict[str, str]]:
+) -> list[ChatCompletionMessageParam]:
     """Assemble the chat `messages` array for a transcript transform or plain prompt.
 
     With a `transcript_id`, the gateway injects the transcript server-side, so we
@@ -88,7 +88,7 @@ def build_messages(
         content = f"{prompt}\n\nTranscript:\n{transcript_text}"
     else:
         content = prompt
-    messages: list[dict[str, str]] = []
+    messages: list[ChatCompletionMessageParam] = []
     if system:
         messages.append({"role": "system", "content": system})
     messages.append({"role": "user", "content": content})
@@ -130,7 +130,7 @@ def complete(
     api_key: str,
     *,
     model: str,
-    messages: list[dict[str, str]],
+    messages: list[ChatCompletionMessageParam],
     max_tokens: int = DEFAULT_MAX_TOKENS,
     transcript_id: str | None = None,
     extra: dict[str, object] | None = None,
@@ -153,7 +153,7 @@ def complete(
     try:
         return client.chat.completions.create(
             model=model,
-            messages=messages,  # type: ignore[arg-type]
+            messages=messages,
             max_tokens=max_tokens,
             extra_body=extra_body or None,
         )
diff --git a/tests/test_llm.py b/tests/test_llm.py
index e86eab3c..8316c7ca 100644
--- a/tests/test_llm.py
+++ b/tests/test_llm.py
@@ -164,7 +164,7 @@ def test_build_messages_transcript_id_uses_tag():
 
 def test_build_messages_inline_text():
     msgs = llm.build_messages("summarize", transcript_text="hello world")
-    assert msgs[0]["content"] == "summarize\n\nTranscript:\nhello world"
+    assert msgs == [{"role": "user", "content": "summarize\n\nTranscript:\nhello world"}]
 
 
 def test_build_messages_with_system_prompt():

From 091ae9f1ebeaa599723e99728d6cb4f4099660b7 Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Sat, 13 Jun 2026 23:03:13 +0000
Subject: [PATCH 2/2] Narrow usage_of return to dict[str, object]
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The token-usage block is only ever emitted as JSON, so its return type
need not be dict[str, Any] — dict[str, object] matches the JSON-payload
convention used throughout the codebase and drops the module's last Any
(and the now-unused typing import).
---
 aai_cli/core/llm.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/aai_cli/core/llm.py b/aai_cli/core/llm.py
index 0c8cb0c7..a10a8052 100644
--- a/aai_cli/core/llm.py
+++ b/aai_cli/core/llm.py
@@ -2,7 +2,7 @@
 
 import json
 from collections.abc import Sequence
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 from aai_cli.core import environments
 from aai_cli.core.errors import APIError, UsageError
@@ -182,7 +182,7 @@ def content_of(response: ChatCompletion) -> str:
     return content or ""
 
 
-def usage_of(response: ChatCompletion) -> dict[str, Any] | None:
+def usage_of(response: ChatCompletion) -> dict[str, object] | None:
     """Return the token-usage block as a plain dict, if present."""
     usage = response.usage
     if usage is None: