fix(runner): handle model-object tool calls in verbose logging

[Sai Asish Y (SAY-5)]

Fixes #54.

DedalusRunner(..., verbose=True).run(..., tools=[...]) crashes with AttributeError on the first response containing tool calls. The verbose print paths call .get("function", {}) on tool-call entries, which works for the ToolCall TypedDicts produced by _extract_tool_calls but not for the SDK model objects read straight off API responses (e.g. in _execute_turns_sync at the Tool calls in response log line, and the message-history dumps).

This adds a small _tool_call_name(tc) helper that accepts either dicts or attribute-style model objects, and uses it in the affected verbose print paths. Non-verbose behavior is unchanged.

Test plan

• New tests/test_runner_verbose.py: unit-tests _tool_call_name for dicts/model objects, and a regression test running runner.run(verbose=True, tools=[add]) against a mocked client returning model-style tool calls (fails on main, passes here).
• pytest tests/test_local_scheduler.py tests/test_runner_verbose.py green.
• ruff check/ruff format --check clean on changed files.

[cursor]

PR Summary

Low Risk
Low risk: changes are confined to verbose-only logging paths and add targeted tests, with no impact to tool execution or API requests when not printing debug output.

Overview
Fixes DedalusRunner crashing in verbose=True mode when tool calls come back as SDK model objects (attribute-only) instead of dicts.

Adds a small _tool_call_name() helper to safely extract tool names from either shape and switches all verbose log sites to use it, plus a regression test that mocks model-style tool calls to ensure run(verbose=True) no longer raises and prints the correct tool name.

^{Reviewed by Cursor Bugbot for commit 69c4276. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Sai Asish Y (SAY-5)]

Good catch. Updated the two remaining verbose sites (the extracted-calls list and the end-of-stream summary) to go through _tool_call_name so model-object tool calls no longer fall back to an inconsistent value.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is ON, but it could not run because the branch was deleted or merged before autofix could start.}

^{Reviewed by Cursor Bugbot for commit 69c4276. Configure here.}

[cursor]

Inline tool-call name logic not refactored to use helper

Low Severity

The new _tool_call_name helper consolidates tool-call name extraction and is used at seven verbose-logging sites, but the equivalent inline implementation in _execute_turns_async (around line 592–601) still duplicates this logic with its own isinstance/getattr branching. This creates an inconsistency where future changes to name-extraction logic would need updating in two places. Additionally, at line 615, the isinstance(tc, dict) / else getattr(tc, "id", "?") branch for tc_id is unreachable because _extract_tool_calls always returns dicts—this adds unnecessary complexity suggesting model objects can appear where they cannot.

Additional Locations (1)

• src/dedalus_labs/lib/runner/core.py#L614-L616

 

^{Reviewed by Cursor Bugbot for commit 69c4276. Configure here.}