[aw][code health] Code quality bundle: toolRequests factory anti-pattern, model_token_map indirection, cost command date-filter duplication

[github-actions]

Overview

Three independent but small code-quality issues that individually are nits but together form a meaningful cleanup. All are in src/copilot_usage/.

---

1. toolRequests uses a misleading default_factory (models.py line 107)

toolRequests: list[dict[str, object]] = Field(
    default_factory=lambda: list[dict[str, object]]()  # ← anti-pattern
)

list[dict[str, object]]() calls the parameterized generic alias as a constructor, which happens to return [] — but this is a confusing side-effect of how Python generics work, not idiomatic code. It's also slower. The correct form is:

toolRequests: list[dict[str, object]] = Field(default_factory=list)

Fix: Replace the lambda with default_factory=list.

---

2. model_token_map unnecessary indirection in build_session_summary (parser.py lines 191, 339–345)

In the active-session code path (no shutdown), model_token_map: dict[str, int] is declared as a variable that at most ever holds one entry, then immediately converted to active_metrics:

model_token_map: dict[str, int] = {}   # declared at top of function
...
if model and total_output_tokens:
    model_token_map[model] = total_output_tokens

active_metrics: dict[str, ModelMetrics] = {}
for m, tokens in model_token_map.items():   # loop over at-most-one-entry dict
    active_metrics[m] = ModelMetrics(...)

The intermediate dict adds cognitive overhead.

Fix: Remove model_token_map and build active_metrics directly:

active_metrics: dict[str, ModelMetrics] = {}
if model and total_output_tokens:
    active_metrics[model] = ModelMetrics(usage=TokenUsage(outputTokens=total_output_tokens))

---

3. cost CLI command inlines date filtering instead of delegating (cli.py lines 395–405)

The summary command delegates date filtering to render_summary, which calls _filter_sessions. The cost command instead manually re-implements filtering inline:

filtered = sessions
if since_aware is not None or until_aware is not None:
    filtered = [
        s for s in sessions
        if s.start_time is not None
        and (since_aware is None or s.start_time >= since_aware)
        and (until_aware is None or s.start_time <= until_aware)
    ]
render_cost_view(filtered)

If _filter_sessions logic ever changes (e.g. timezone handling, edge cases), the cost path won't get the update automatically.

Fix: Add since: datetime | None = None, until: datetime | None = None parameters to render_cost_view (in report.py), call _filter_sessions internally (same as render_summary does), and remove the inline filtering from cli.py::cost.

---

Testing requirement

• Item 1: No additional tests needed; existing tests cover the field. Verify with pytest tests/copilot_usage/test_models.py -v.
• Item 2: Add a unit test in test_parser.py for an active session with a known model + token count that asserts active_metrics is set correctly without intermediate dict.
• Item 3: Add a test in test_cli.py (or test_report.py) that passes since/until directly to render_cost_view and verifies filtering matches _filter_sessions behaviour; also verify the cost CLI command still filters correctly via the Click runner.

Generated by Code Health Analysis · ◷