fix: resolve all mypy type errors in CLI files

Add missing type annotations to benchmark.py context managers, event
handlers, LoadedCases iteration methods, and fix union-attr on BaseLLM.
Fix no-any-return errors in agent_tui.py and make action_quit async to
match the Textual App supertype. Add type annotations to
_BenchmarkLiveProgress methods in cli.py and fix icon redefinition.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

lib/cli/src/crewai_cli/agent_tui.py

@@ -139,7 +139,7 @@ def _load_config(base: Path) -> dict[str, Any]:
         return {"rooms": {"common": {"agents": [], "engagement": "organic"}}}
     try:
         raw = config_path.read_text(encoding="utf-8")
-        return json.loads(_strip_jsonc(raw))
+        return json.loads(_strip_jsonc(raw))  # type: ignore[no-any-return]
     except Exception:
         return {"rooms": {"common": {"agents": [], "engagement": "organic"}}}
 
@@ -559,12 +559,12 @@ class AgentTUI(App[None]):
 
     def _room_engagement(self, room_key: str) -> str:
         if room_key in self._rooms:
-            return self._rooms[room_key].get("engagement", "organic")
+            return str(self._rooms[room_key].get("engagement", "organic"))
         return "organic"
 
     def _room_agents(self, room_key: str) -> list[str]:
         if room_key in self._rooms:
-            return self._rooms[room_key].get("agents", self._agent_names[:])
+            return list(self._rooms[room_key].get("agents", self._agent_names[:]))
         return self._agent_names[:]
 
     def on_mount(self) -> None:
@@ -1679,7 +1679,7 @@ class AgentTUI(App[None]):
 
     # ── Actions ──
 
-    def action_quit(self) -> None:
+    async def action_quit(self) -> None:
         """Graceful shutdown: stop scheduler, silence event bus, then exit."""
         self._mount_sys("Shutting down...")
         if self._scheduler:

lib/cli/src/crewai_cli/benchmark.py

@@ -3,7 +3,7 @@
 from __future__ import annotations
 
 import asyncio
-from collections.abc import Callable
+from collections.abc import Callable, Iterator
 import json
 from pathlib import Path
 import re
@@ -49,10 +49,10 @@ class LoadedCases:
     def __len__(self) -> int:
         return len(self.cases)
 
-    def __iter__(self):
+    def __iter__(self) -> Iterator[BenchmarkCase]:
         return iter(self.cases)
 
-    def __getitem__(self, index):
+    def __getitem__(self, index: int) -> BenchmarkCase:
         return self.cases[index]
 
 
@@ -136,6 +136,7 @@ async def _judge_with_llm(
     from crewai.utilities.llm_utils import create_llm
 
     judge_llm = create_llm(judge_model)
+    assert judge_llm is not None
 
     prompt = (
         "You are an evaluation judge. Score the following response on a scale of 0.0 to 1.0.\n\n"
@@ -444,7 +445,7 @@ async def run_benchmark(
 class SuppressBenchmarkOutput:
     """Context manager that silences console formatter and noisy logging during benchmarks."""
 
-    def __enter__(self):
+    def __enter__(self) -> "SuppressBenchmarkOutput":
         import logging
 
         self._saved_formatter = None
@@ -453,7 +454,7 @@ class SuppressBenchmarkOutput:
                 TraceCollectionListener,
             )
 
-            listener = TraceCollectionListener._instance
+            listener = TraceCollectionListener._instance  # type: ignore[misc]
             if listener:
                 self._saved_formatter = listener.formatter
                 listener.formatter = None
@@ -471,7 +472,7 @@ class SuppressBenchmarkOutput:
             lg.setLevel(logging.CRITICAL)
         return self
 
-    def __exit__(self, *exc):
+    def __exit__(self, *exc: object) -> None:
         for lg, level in self._loggers:
             lg.setLevel(level)
         if self._saved_formatter is not None:
@@ -480,7 +481,7 @@ class SuppressBenchmarkOutput:
                     TraceCollectionListener,
                 )
 
-                listener = TraceCollectionListener._instance
+                listener = TraceCollectionListener._instance  # type: ignore[misc]
                 if listener:
                     listener.formatter = self._saved_formatter
             except Exception:
@@ -490,7 +491,7 @@ class SuppressBenchmarkOutput:
 class VerboseBenchmarkOutput:
     """Context manager that subscribes to NewAgent events and prints them for debugging."""
 
-    def __enter__(self):
+    def __enter__(self) -> "VerboseBenchmarkOutput":
         import logging
         import sys
 
@@ -513,7 +514,7 @@ class VerboseBenchmarkOutput:
                 TraceCollectionListener,
             )
 
-            listener = TraceCollectionListener._instance
+            listener = TraceCollectionListener._instance  # type: ignore[misc]
             if listener:
                 self._saved_formatter = listener.formatter
                 listener.formatter = None
@@ -537,38 +538,38 @@ class VerboseBenchmarkOutput:
         w = sys.stderr.write
         fl = sys.stderr.flush
 
-        def _on_llm_start(_src, ev: NewAgentLLMCallStartedEvent):
+        def _on_llm_start(_src: Any, ev: NewAgentLLMCallStartedEvent) -> None:
             w(f"\033[36m[llm] calling {ev.model}…\033[0m\n")
             fl()
 
-        def _on_llm_done(_src, ev: NewAgentLLMCallCompletedEvent):
+        def _on_llm_done(_src: Any, ev: NewAgentLLMCallCompletedEvent) -> None:
             w(
                 f"\033[36m[llm] {ev.model}  {ev.input_tokens}→{ev.output_tokens} tokens  {ev.response_time_ms}ms\033[0m\n"
             )
             fl()
 
-        def _on_llm_fail(_src, ev: NewAgentLLMCallFailedEvent):
+        def _on_llm_fail(_src: Any, ev: NewAgentLLMCallFailedEvent) -> None:
             w(f"\033[31m[llm] FAILED: {ev.error[:200]}\033[0m\n")
             fl()
 
-        def _on_tool_start(_src, ev: NewAgentToolUsageStartedEvent):
+        def _on_tool_start(_src: Any, ev: NewAgentToolUsageStartedEvent) -> None:
             w(f"\033[33m[tool] using {ev.tool_name}…\033[0m\n")
             fl()
 
-        def _on_tool_done(_src, ev: NewAgentToolUsageCompletedEvent):
+        def _on_tool_done(_src: Any, ev: NewAgentToolUsageCompletedEvent) -> None:
             w(f"\033[33m[tool] {ev.tool_name} done\033[0m\n")
             fl()
 
-        def _on_tool_fail(_src, ev: NewAgentToolUsageFailedEvent):
+        def _on_tool_fail(_src: Any, ev: NewAgentToolUsageFailedEvent) -> None:
             w(f"\033[31m[tool] {ev.tool_name} FAILED: {ev.error[:200]}\033[0m\n")
             fl()
 
-        def _on_status(_src, ev: NewAgentStatusUpdateEvent):
+        def _on_status(_src: Any, ev: NewAgentStatusUpdateEvent) -> None:
             if ev.detail:
                 w(f"\033[2m[status] {ev.state}: {ev.detail}\033[0m\n")
                 fl()
 
-        def _on_summarized(_src, ev: NewAgentContextSummarizedEvent):
+        def _on_summarized(_src: Any, ev: NewAgentContextSummarizedEvent) -> None:
             w("\033[35m[context] summarized — context was too large\033[0m\n")
             fl()
 
@@ -583,14 +584,14 @@ class VerboseBenchmarkOutput:
             (NewAgentContextSummarizedEvent, _on_summarized),
         ]
         for event_type, handler in pairs:
-            self._bus.on(event_type)(handler)
+            self._bus.on(event_type)(handler)  # type: ignore[arg-type]
             self._handlers.append((event_type, handler))
         return self
 
-    def __exit__(self, *exc):
+    def __exit__(self, *exc: object) -> None:
         for event_type, handler in self._handlers:
             try:
-                self._bus.off(event_type, handler)
+                self._bus.off(event_type, handler)  # type: ignore[arg-type]
             except Exception:
                 pass
         for lg, level in self._loggers:
@@ -601,7 +602,7 @@ class VerboseBenchmarkOutput:
                     TraceCollectionListener,
                 )
 
-                listener = TraceCollectionListener._instance
+                listener = TraceCollectionListener._instance  # type: ignore[misc]
                 if listener:
                     listener.formatter = self._saved_formatter
             except Exception:
@@ -619,7 +620,7 @@ class ArtifactsSandbox:
         self._base = Path(base)
         self._prev_cwd: str | None = None
 
-    def __enter__(self):
+    def __enter__(self) -> "ArtifactsSandbox":
         import os
 
         self._base.mkdir(parents=True, exist_ok=True)
@@ -630,7 +631,7 @@ class ArtifactsSandbox:
         os.chdir(self._base)
         return self
 
-    def __exit__(self, *exc):
+    def __exit__(self, *exc: object) -> None:
         import os
 
         if self._prev_cwd:

lib/cli/src/crewai_cli/cli.py

@@ -198,7 +198,7 @@ def train(n_iterations: int, filename: str) -> None:
         train_crew(n_iterations, filename)
 
 
-def _train_new_agents(agent_files: list, n_iterations: int) -> None:
+def _train_new_agents(agent_files: list[Any], n_iterations: int) -> None:
     """Run interactive training for NewAgent agents.
 
     For each agent, loads benchmark cases, runs them, shows the response,
@@ -631,14 +631,14 @@ def _read_config(*keys: str) -> Any:
 class _BenchmarkLiveProgress:
     """Live parallel progress display for benchmark runs."""
 
-    def __init__(self, console=None):
+    def __init__(self, console: Any = None) -> None:
         from rich.console import Console
 
         self._console = console or Console()
-        self._state: dict[str, dict] = {}
-        self._live = None
+        self._state: dict[str, dict[str, Any]] = {}
+        self._live: Any = None
 
-    def start(self):
+    def start(self) -> None:
         from rich.live import Live
 
         self._live = Live(
@@ -649,13 +649,13 @@ class _BenchmarkLiveProgress:
         )
         self._live.start()
 
-    def stop(self):
+    def stop(self) -> None:
         if self._live:
             self._live.update(self._render())
             self._live.stop()
             self._live = None
 
-    def on_progress(self, event: dict) -> None:
+    def on_progress(self, event: dict[str, Any]) -> None:
         t = event["type"]
         model = event.get("model", "")
 
@@ -695,7 +695,7 @@ class _BenchmarkLiveProgress:
         if self._live:
             self._live.update(self._render())
 
-    def _render(self):
+    def _render(self) -> Any:
         from rich import box
         from rich.spinner import Spinner
         from rich.table import Table
@@ -721,6 +721,7 @@ class _BenchmarkLiveProgress:
             table.add_column("", no_wrap=True, justify="right")  # cost
 
         for model, info in self._state.items():
+            icon: Any
             if info["status"] == "done":
                 icon = Text("✓", style="green")
                 color = _score_color(info["avg"])
@@ -758,7 +759,7 @@ class _BenchmarkLiveProgress:
 
 
 def _test_new_agents(
-    agent_files: list,
+    agent_files: list[Any],
     n_iterations: int,
     model: str | None,
     threshold: float,
@@ -782,7 +783,7 @@ def _test_new_agents(
         tests_dir = Path("benchmarks")
 
     # Collect valid agents + cases
-    jobs: list[dict] = []
+    jobs: list[dict[str, Any]] = []
     for agent_path in agent_files:
         agent_name = agent_path.stem
         cases_path = tests_dir / f"{agent_name}_cases.json"
@@ -816,8 +817,8 @@ def _test_new_agents(
     # Progress display — prefix model key with agent name
     progress = None if verbose else _BenchmarkLiveProgress(console=_con)
 
-    def _make_progress_cb(agent_name: str):
-        def _cb(event: dict) -> None:
+    def _make_progress_cb(agent_name: str) -> Any:
+        def _cb(event: dict[str, Any]) -> None:
             if progress is not None:
                 prefixed = dict(event)
                 if "model" in prefixed:
@@ -826,7 +827,7 @@ def _test_new_agents(
 
         return _cb
 
-    async def _run_all():
+    async def _run_all() -> Any:
         tasks = []
         for job in jobs:
             tasks.append(
@@ -858,6 +859,7 @@ def _test_new_agents(
     )
 
     if not verbose:
+        assert progress is not None
         progress.start()
     try:
         with ArtifactsSandbox():
@@ -869,6 +871,7 @@ def _test_new_agents(
                     all_results = asyncio.run(_run_all())
     finally:
         if not verbose:
+            assert progress is not None
             progress.stop()
 
     # Evaluate results