Fix six issues found running the agent examples

- settings.py docstring claimed the default model was
  anthropic/claude-sonnet-4-6; the code defaults to openai/gpt-4o (#485).
- Add langchain-anthropic to the langchain and agents extras, so
  langgraph/43_react_agent_multi_model.py runs after the documented
  install. It requires anthropic >=0.120 and langchain-core >=1.5.4, which
  the lock file now reflects (#486).
- Wrap the body of langgraph/46_crash_and_resume.py in a __main__ guard.
  Without it, spawned workers re-imported the module and re-ran the whole
  demo. The tools and graph stay at module level so workers can still
  rebuild them (#489).
- Replace anthropic/claude-sonnet-4-20250514, which the API answers with
  404 not_found_error, with anthropic/claude-sonnet-5, the replacement
  #494 verified against a live server (#480).
- 74_cli_error_output.py asserted against result.output, which is a dict,
  so the substring check tested keys and the example could never pass. Use
  str(result.output), as the other examples that need a string do (#481).
- Point run_examples.sh at examples/agents. It globbed examples/, where no
  numbered examples live, so it collected nothing and always passed (#484).

Because that glob matched nothing, the script's skip list had never been
exercised. Add the examples that must not run unattended: seven that act on
real third-party accounts, and eight that block on stdin with no canned
response. --all still runs them.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: ambiorix2099

examples/agentic_workflows/mcp_weather_agent.py

@@ -90,7 +90,7 @@ def create_mcp_agent_workflow(executor: WorkflowExecutor, mcp_server: str) -> Co
     plan_task = LlmChatComplete(
         task_ref_name="plan_action",
         llm_provider="anthropic",
-        model="claude-sonnet-4-20250514",
+        model="claude-sonnet-5",
         messages=[
             ChatMessage(
                 role="system",

examples/agents/56_rag_agent.py

@@ -65,7 +65,7 @@
         "text": (
             "Agent Configuration. Agents are defined with a name, model, instructions, "
             "and tools. The model field uses the format 'provider/model_name', e.g. "
-            "'openai/gpt-4o' or 'anthropic/claude-sonnet-4-20250514'. Instructions can be "
+            "'openai/gpt-4o' or 'anthropic/claude-sonnet-5'. Instructions can be "
             "a string or a PromptTemplate referencing a stored prompt. Tools can be "
             "@tool-decorated Python functions, http_tool for REST APIs, mcp_tool for "
             "MCP servers, or agent_tool to wrap another agent as a callable tool. "

examples/agents/59_coding_agent.py

@@ -26,7 +26,7 @@
 
 qa_tester = Agent(
     name="qa_tester",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a meticulous QA engineer. Review the code written by the "
         "coder for correctness, edge cases, and bugs. Write and execute test "
@@ -45,7 +45,7 @@
 
 coder = Agent(
     name="coder",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are an expert Python developer. Write clean, well-structured "
         "Python code to solve the given problem. Always execute your code to "

examples/agents/60_github_coding_agent.py

@@ -249,7 +249,7 @@ def create_pull_request(title: str, body: str, issue_number: int = 0) -> str:
 
 github_agent = Agent(
     name="github_agent",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a GitHub operations specialist. You handle all git and "
         "GitHub CLI interactions.\n\n"
@@ -282,7 +282,7 @@ def create_pull_request(title: str, body: str, issue_number: int = 0) -> str:
 
 coder = Agent(
     name="coder",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are an expert developer. Write clean, well-structured code.\n\n"
         "WHEN YOU RECEIVE A TASK:\n"
@@ -311,7 +311,7 @@ def create_pull_request(title: str, body: str, issue_number: int = 0) -> str:
 
 qa_tester = Agent(
     name="qa_tester",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a meticulous QA engineer. Review the code written by the "
         "coder for correctness, edge cases, and bugs.\n\n"
@@ -340,7 +340,7 @@ def create_pull_request(title: str, body: str, issue_number: int = 0) -> str:
 
 coding_team = Agent(
     name="coding_team",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a coding team coordinator. Delegate the incoming request "
         "to github_agent to get started — it will pick an issue and set "

examples/agents/60a_github_coding_agent_simple.py

@@ -41,7 +41,7 @@
 
 github_agent = Agent(
     name="github_agent",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a GitHub operations specialist. You handle all git and "
         "GitHub CLI interactions.\n\n"
@@ -77,7 +77,7 @@
 
 coder = Agent(
     name="coder",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are an expert developer. You write clean, well-structured code.\n\n"
         f"The repo is cloned at {WORK_DIR}.\n\n"
@@ -109,7 +109,7 @@
 
 qa_tester = Agent(
     name="qa_tester",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a meticulous QA engineer. Review the code written by the "
         "coder for correctness, edge cases, and bugs.\n\n"
@@ -134,7 +134,7 @@
 
 coding_team = Agent(
     name="coding_team",
-    model="anthropic/claude-sonnet-4-20250514",
+    model="anthropic/claude-sonnet-5",
     instructions=(
         "You are a coding team coordinator. Delegate the incoming request "
         "to github_agent to get started — it will pick an issue and set "

examples/agents/74_cli_error_output.py

@@ -36,7 +36,7 @@
     with AgentRuntime() as rt:
         result = rt.run(agent, prompt)
         result.print_result()
-        output = result.output or ""
+        output = str(result.output)
 
         # Verify the agent saw the error output
         assert "No such file or directory" in output or "nonexistent" in output, (

examples/agents/adk/35_rag_agent.py

@@ -76,7 +76,7 @@
         "text": (
             "Agent Configuration. Agents are defined with a name, model, instructions, "
             "and tools. The model field uses the format 'provider/model_name', e.g. "
-            "'openai/gpt-4o' or 'anthropic/claude-sonnet-4-20250514'. Instructions can be "
+            "'openai/gpt-4o' or 'anthropic/claude-sonnet-5'. Instructions can be "
             "a string or a PromptTemplate referencing a stored prompt. Tools can be "
             "@tool-decorated Python functions, http_tool for REST APIs, mcp_tool for "
             "MCP servers, or agent_tool to wrap another agent as a callable tool. "

examples/agents/langgraph/46_crash_and_resume.py

@@ -104,80 +104,80 @@ def generate_report(analysis: str) -> str:
 )
 
 
-# -- Phase 1: Deploy, start, serve briefly, then crash --------------------
-
-print("=" * 60)
-print("Phase 1: Deploy + start, then simulate crash")
-print("=" * 60)
-
-with AgentRuntime() as runtime:
-    # Deploy the workflow definition (in production, do this once in CI/CD)
-    runtime.deploy(graph)
-    print("Agent deployed to server.")
-
-    # Start an execution by name — the agent is already deployed on the server,
-    # so we only need to send the name and prompt (not the full graph object).
-    handle = runtime.start(
-        "sales_analyst",
-        "Fetch the Q4 2025 sales data, run a full trend analysis on it, "
-        "then generate an executive summary report. "
-        "Call each tool in sequence — do not skip any step.",
-    )
-    print(f"Execution started: {handle.execution_id}")
-
-    # Save execution_id so we can check status later
-    with open(SESSION_FILE, "w") as f:
-        f.write(handle.execution_id)
-
-    # Serve workers just long enough for the first tool to start
-    print("\nServing workers briefly...")
-    runtime.serve(graph, blocking=False)
-    time.sleep(8)
-
-print("\nRuntime closed — workers are dead, workflow persists on server.")
-print()
-
-with open(SESSION_FILE) as f:
-    saved_execution_id = f.read().strip()
-
-# -- Pause: let the user see the stalled execution in the UI --------------
-
-ui_link = f"{UI_BASE}/execution/{saved_execution_id}"
-print("-" * 60)
-print("Open the Conductor UI to see the execution in RUNNING state:")
-print(f"  {ui_link}")
-print()
-print("The workflow is alive on the server but stalled — no workers are")
-print("polling to pick up the next task.  The completed steps are")
-print("preserved; only the remaining steps need to run.")
-print("-" * 60)
-input("\nPress Enter to resume (restart workers)...")
-print()
-
-
-# -- Phase 2: Restart serve — workers pick up stalled tasks ----------------
-
-print("=" * 60)
-print("Phase 2: Restart serve() — workers reconnect automatically")
-print("=" * 60)
-
-with AgentRuntime() as runtime:
-    # serve() re-registers the same workers.  The server dispatches
-    # stalled tasks to them — no resume() call needed.
-    print("\nServing workers (non-blocking for demo)...")
-    runtime.serve(graph, blocking=False)
-
-    # Poll until the execution completes
-    print(f"Polling execution: {saved_execution_id}")
-    status = runtime.get_status(saved_execution_id)
-    while not status.is_complete:
-        time.sleep(2)
+if __name__ == "__main__":
+    # -- Phase 1: Deploy, start, serve briefly, then crash --------------------
+
+    print("=" * 60)
+    print("Phase 1: Deploy + start, then simulate crash")
+    print("=" * 60)
+
+    with AgentRuntime() as runtime:
+        # Deploy the workflow definition (in production, do this once in CI/CD)
+        runtime.deploy(graph)
+        print("Agent deployed to server.")
+
+        # Start an execution by name — the agent is already deployed on the server,
+        # so we only need to send the name and prompt (not the full graph object).
+        handle = runtime.start(
+            "sales_analyst",
+            "Fetch the Q4 2025 sales data, run a full trend analysis on it, "
+            "then generate an executive summary report. "
+            "Call each tool in sequence — do not skip any step.",
+        )
+        print(f"Execution started: {handle.execution_id}")
+
+        # Save execution_id so we can check status later
+        with open(SESSION_FILE, "w") as f:
+            f.write(handle.execution_id)
+
+        # Serve workers just long enough for the first tool to start
+        print("\nServing workers briefly...")
+        runtime.serve(graph, blocking=False)
+        time.sleep(8)
+
+    print("\nRuntime closed — workers are dead, workflow persists on server.")
+    print()
+
+    with open(SESSION_FILE) as f:
+        saved_execution_id = f.read().strip()
+
+    # -- Pause: let the user see the stalled execution in the UI --------------
+
+    ui_link = f"{UI_BASE}/execution/{saved_execution_id}"
+    print("-" * 60)
+    print("Open the Conductor UI to see the execution in RUNNING state:")
+    print(f"  {ui_link}")
+    print()
+    print("The workflow is alive on the server but stalled — no workers are")
+    print("polling to pick up the next task.  The completed steps are")
+    print("preserved; only the remaining steps need to run.")
+    print("-" * 60)
+    input("\nPress Enter to resume (restart workers)...")
+    print()
+
+    # -- Phase 2: Restart serve — workers pick up stalled tasks ----------------
+
+    print("=" * 60)
+    print("Phase 2: Restart serve() — workers reconnect automatically")
+    print("=" * 60)
+
+    with AgentRuntime() as runtime:
+        # serve() re-registers the same workers.  The server dispatches
+        # stalled tasks to them — no resume() call needed.
+        print("\nServing workers (non-blocking for demo)...")
+        runtime.serve(graph, blocking=False)
+
+        # Poll until the execution completes
+        print(f"Polling execution: {saved_execution_id}")
         status = runtime.get_status(saved_execution_id)
-        print(f"  status: {status.status}")
+        while not status.is_complete:
+            time.sleep(2)
+            status = runtime.get_status(saved_execution_id)
+            print(f"  status: {status.status}")
 
-    print(f"\nStatus: {status.status}")
-    print(f"Output: {status.output}")
-    print("\nCheck the completed execution in the UI:")
-    print(f"  {ui_link}")
+        print(f"\nStatus: {status.status}")
+        print(f"Output: {status.output}")
+        print("\nCheck the completed execution in the UI:")
+        print(f"  {ui_link}")
 
-print("\nDone — same workflow, seamless resume after simulated crash.")
+    print("\nDone — same workflow, seamless resume after simulated crash.")

examples/agents/settings.py

@@ -3,10 +3,10 @@
 Set ``CONDUCTOR_AGENT_LLM_MODEL`` as an environment variable to override the
 default model used by all examples::
 
-    export CONDUCTOR_AGENT_LLM_MODEL=anthropic/claude-sonnet-4-20250514
+    export CONDUCTOR_AGENT_LLM_MODEL=anthropic/claude-sonnet-5
     export CONDUCTOR_AGENT_LLM_MODEL=google_gemini/gemini-2.0-flash
 
-If unset, defaults to ``anthropic/claude-sonnet-4-6``.
+If unset, defaults to ``openai/gpt-4o``.
 
 ``CONDUCTOR_AGENT_SECONDARY_LLM_MODEL`` provides a second model for multi-model examples
 (e.g., cheap triage vs capable specialist). Defaults to ``openai/gpt-4o``.