fix: add --dangerously-skip-permissions + clean up logging (#31)

- Add --dangerously-skip-permissions to Claude CLI (fixes permission prompt hang)
- Clean up logfire calls with proper kwargs
- Remove duplicate/broken print statements

Author: nikhilwoodruff

src/policyengine_api/agent_sandbox.py

@@ -136,6 +136,7 @@ def run_claude_code_in_sandbox(
     )
 
     # Run Claude Code with the question
+    # --dangerously-skip-permissions: auto-accept permission prompts (required for non-interactive)
     logfire.info("run_claude_code_in_sandbox: starting claude CLI")
     process = sb.exec(
         "claude",
@@ -144,6 +145,7 @@ def run_claude_code_in_sandbox(
         "--output-format",
         "stream-json",
         "--verbose",
+        "--dangerously-skip-permissions",
         "--allowedTools",
         "mcp__policyengine__*,Bash,Read,Grep,Glob,Write,Edit",
     )

src/policyengine_api/api/agent.py

@@ -10,6 +10,7 @@
 import os
 from uuid import uuid4
 
+import logfire
 from fastapi import APIRouter, HTTPException
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
@@ -88,32 +89,24 @@ async def _stream_modal_sandbox(question: str, api_base_url: str):
     """Stream output from Claude Code running in Modal Sandbox."""
     from concurrent.futures import ThreadPoolExecutor
 
-    import logfire
-
-    # Immediate yield to confirm generator is running
-    print(f"[AGENT] _stream_modal_sandbox called with question: {question[:50]}")
-    yield f"data: {json.dumps({'type': 'output', 'content': '[DEBUG] Generator started\\n'})}\n\n"
+    # Immediate log and yield
+    logfire.info("_stream_modal_sandbox: started", question=question[:100])
 
     sb = None
     executor = ThreadPoolExecutor(max_workers=1)
     try:
         from policyengine_api.agent_sandbox import run_claude_code_in_sandbox
 
-        print("[AGENT] About to call run_claude_code_in_sandbox")
         logfire.info(
-            "Creating Modal sandbox", question=question[:100], api_base_url=api_base_url
+            "_stream_modal_sandbox: creating sandbox", api_base_url=api_base_url
         )
-        yield f"data: {json.dumps({'type': 'output', 'content': '[DEBUG] Creating Modal sandbox...\\n'})}\n\n"
 
         # Run blocking Modal SDK calls in thread pool to avoid blocking event loop
         loop = asyncio.get_event_loop()
-        print("[AGENT] Calling run_in_executor for sandbox creation")
         sb, process = await loop.run_in_executor(
             executor, run_claude_code_in_sandbox, question, api_base_url
         )
-        print("[AGENT] Sandbox created successfully")
-        logfire.info("Modal sandbox created, streaming output")
-        yield f"data: {json.dumps({'type': 'output', 'content': '[DEBUG] Sandbox created, starting stream...\\n'})}\n\n"
+        logfire.info("_stream_modal_sandbox: sandbox created")
 
         # Poll for lines with timeout to allow other async tasks
         import queue
@@ -190,11 +183,7 @@ def stream_reader():
                 raise
 
     except Exception as e:
-        import traceback
-
-        tb = traceback.format_exc()
-        print(f"[AGENT] Exception in _stream_modal_sandbox: {e}\n{tb}")
-        logfire.exception("Modal sandbox failed", error=str(e))
+        logfire.exception("_stream_modal_sandbox: failed", error=str(e))
         yield f"data: {json.dumps({'type': 'error', 'content': f'Sandbox error: {str(e)}'})}\n\n"
         yield f"data: {json.dumps({'type': 'done', 'returncode': 1})}\n\n"
     finally:
@@ -226,19 +215,20 @@ async def stream_analysis(request: AskRequest):
     data: {"type": "done", "returncode": 0}
     ```
     """
-    print(f"[AGENT] /stream endpoint called with question: {request.question[:50]}")
-    print(f"[AGENT] agent_use_modal = {settings.agent_use_modal}")
     api_base_url = settings.policyengine_api_url
-    print(f"[AGENT] api_base_url = {api_base_url}")
+    logfire.info(
+        "stream_analysis: called",
+        question=request.question[:100],
+        agent_use_modal=settings.agent_use_modal,
+        api_base_url=api_base_url,
+    )
 
     if settings.agent_use_modal:
-        print("[AGENT] Using Modal sandbox path")
         return StreamingResponse(
             _stream_modal_sandbox(request.question, api_base_url),
             media_type="text/event-stream",
         )
     else:
-        print("[AGENT] Using local claude path")
         return StreamingResponse(
             _stream_claude_code(request.question, api_base_url),
             media_type="text/event-stream",