chore: make the tool-selection eval tolerant of Gemini's search-first pick (#95)

* chore: I hate em dash

* chore: make the eval more tolerent

* Update test_tool_selection.py

* fix typo

Author: WilliamBergamin

tests/eval/test_tool_selection.py

@@ -107,17 +107,17 @@ class ToolChoice(BaseModel):
     {
         "id": "ambiguous-add-reaction-releases",
         "prompt": "Add a :tada: reaction to the latest message in #releases",
-        "accepted_tools": ["slack_add_reaction", "slack_read_channel"],
+        "accepted_tools": ["slack_add_reaction", "slack_read_channel", "slack_search_public", "slack_search_channels"],
     },
     {
         "id": "ambiguous-reply-in-thread",
         "prompt": "Reply 'we're on it' in the thread on the outage message in #incidents",
-        "accepted_tools": ["slack_send_message", "slack_read_thread"],
+        "accepted_tools": ["slack_send_message", "slack_read_thread", "slack_search_public", "slack_search_channels"],
     },
     {
         "id": "ambiguous-read-thread-replies",
         "prompt": "Show me all the replies in the thread on the latest message in #support",
-        "accepted_tools": ["slack_read_thread", "slack_read_channel"],
+        "accepted_tools": ["slack_read_thread", "slack_read_channel", "slack_search_public", "slack_search_channels"],
     },
     {
         "id": "ambiguous-lookup-user-by-email",
@@ -177,7 +177,7 @@ def build_prompt(available_tools: list[ToolCall], prompt: str) -> str:
 User request: {prompt}
 
 Pick the single tool that performs the action the user is asking for. Any channel name,
-channel ID, or user ID already in the request is usable as-is — do not pick a search tool
+channel ID, or user ID already in the request is usable as-is. Do not pick a search tool
 just to resolve it into an ID first. Respond with the tool's exact name."""