closed — replaced by #4

scripts/claude_worker.py

@@ -32,6 +32,8 @@
 
 import httpx
 
+from multihead.subprocess_utils import no_window_flags
+
 logging.basicConfig(
     level=logging.INFO,
     format="[worker] %(asctime)s %(levelname)s %(message)s",
@@ -376,6 +378,7 @@ def _invoke_claude(self, prompt: str, session_id: str | None = None) -> dict:
                 cwd=CLAUDE_WORK_DIR,
                 timeout=SUBPROCESS_TIMEOUT,
                 env=env,
+                creationflags=no_window_flags(),
             )
         except subprocess.TimeoutExpired:
             elapsed = time.monotonic() - start
@@ -458,6 +461,7 @@ def _wake_interactive(self, message: str) -> dict:
             check = subprocess.run(
                 ["tmux", "has-session", "-t", self.tmux_target],
                 capture_output=True, timeout=5,
+                creationflags=no_window_flags(),
             )
             if check.returncode != 0:
                 return {"error": f"tmux session '{self.tmux_target}' not found. "
@@ -467,6 +471,7 @@ def _wake_interactive(self, message: str) -> dict:
             subprocess.run(
                 ["tmux", "send-keys", "-t", self.tmux_target, message, "Enter"],
                 check=True, timeout=5,
+                creationflags=no_window_flags(),
             )
             return {"result": f"Sent to tmux session '{self.tmux_target}'", "mode": "interactive"}
 

src/multihead/adapters/claude_agent_sdk.py

@@ -12,6 +12,7 @@
 
 import logging
 import os
+import sys
 from collections.abc import AsyncIterator
 from typing import Any
 
@@ -62,7 +63,7 @@ class ClaudeAgentSDKAdapter(HeadAdapter):
     for typed messages, streaming, session management, and hooks.
 
     Configuration via manifest.extra:
-        model: str          — Claude model (default: from env or "claude-sonnet-4-20250514")
+        model: str          — Claude model (default: from env or "claude-sonnet-4-6")
         max_turns: int      — Max agent turns per call (default: 25)
         max_budget_usd: float — Budget cap per call (default: 5.0)
         permission_mode: str — "bypassPermissions" | "acceptEdits" | "default"
@@ -72,12 +73,40 @@ class ClaudeAgentSDKAdapter(HeadAdapter):
         effort: str         — "low" | "medium" | "high" | "max"
     """
 
+    _anyio_patched = False
+
+    @classmethod
+    def _patch_anyio(cls) -> None:
+        """Monkey-patch anyio.open_process to suppress console windows on Windows.
+
+        The Claude Agent SDK spawns subprocesses via anyio, which does not pass
+        CREATE_NO_WINDOW by default. This causes popup cmd windows during Night
+        Shift and other headless operations. Only patches once per process.
+        """
+        if cls._anyio_patched or sys.platform != "win32":
+            return
+        try:
+            import anyio as _anyio
+            import subprocess as _subprocess
+
+            _original = _anyio.open_process
+
+            async def _patched(*args: Any, **kwargs: Any) -> Any:
+                kwargs.setdefault("creationflags", _subprocess.CREATE_NO_WINDOW)
+                return await _original(*args, **kwargs)
+
+            _anyio.open_process = _patched  # type: ignore[assignment]
+            cls._anyio_patched = True
+        except ImportError:
+            pass  # anyio not installed — SDK won't work either
+
     def __init__(self, manifest: HeadManifest) -> None:
         super().__init__(manifest)
+        self._patch_anyio()
         extra = manifest.extra or {}
 
         self._model = extra.get("model") or manifest.model or os.environ.get(
-            "CLAUDE_MODEL", "claude-opus-4-6"
+            "CLAUDE_MODEL", "claude-sonnet-4-6"
         )
         self._max_turns = extra.get("max_turns", _DEFAULT_MAX_TURNS)
         self._max_budget = extra.get("max_budget_usd", _DEFAULT_MAX_BUDGET_USD)

src/multihead/api/routes_nightshift.py

@@ -7,6 +7,9 @@
 from typing import Any
 
 from fastapi import APIRouter, BackgroundTasks, Query, Request
+from fastapi.responses import JSONResponse
+
+from multihead.models import HeadState
 
 router = APIRouter()
 
@@ -16,7 +19,7 @@
     "running": False,
     "last_report": None,
     "current_stage": None,
-    "progress": deque(maxlen=100),
+    "progress": deque(maxlen=5000),
 }
 
 
@@ -28,12 +31,25 @@ async def trigger_nightshift(
     concurrency: int = Query(1, description="Parallel LLM calls per stage (1=sequential)"),
 ) -> dict[str, Any]:
     """Trigger a Night Shift run in the background."""
+    # Check for at least one active head before starting
+    head_manager = request.app.state.head_manager
+    states = head_manager.get_states()
+    has_active = any(
+        info.get("state") == HeadState.ACTIVE.value
+        for info in states.values()
+    )
+    if not has_active:
+        return JSONResponse(
+            status_code=400,
+            content={"error": "No active heads — wake a head before running Night Shift"},
+        )
+
     async with _nightshift_lock:
         if _nightshift_status["running"]:
             return {"status": "already_running"}
         _nightshift_status["running"] = True
         _nightshift_status["current_stage"] = None
-        _nightshift_status["progress"] = deque(maxlen=100)
+        _nightshift_status["progress"] = deque(maxlen=5000)
 
     night_shift = request.app.state.night_shift
     night_shift.config.concurrency = max(1, concurrency)

src/multihead/api/routes_system.py

@@ -12,6 +12,8 @@
 
 from fastapi import APIRouter, Query, Request
 
+from multihead.subprocess_utils import no_window_flags
+
 logger = logging.getLogger(__name__)
 
 router = APIRouter()
@@ -62,7 +64,7 @@ async def restart():
     project_root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
     # Spawn a new server process directly (bypasses mh.cmd issues with nested shells)
     flags = (
-        subprocess.DETACHED_PROCESS | subprocess.CREATE_NEW_PROCESS_GROUP
+        subprocess.DETACHED_PROCESS | subprocess.CREATE_NEW_PROCESS_GROUP | no_window_flags()
         if sys.platform == "win32"
         else 0
     )

src/multihead/autonomous_executor/strategies.py

@@ -12,6 +12,7 @@
 from abc import ABC, abstractmethod
 from typing import Any
 
+from multihead.subprocess_utils import no_window_flags
 from .models import StepExecutionResult
 
 logger = logging.getLogger(__name__)
@@ -213,6 +214,7 @@ async def execute_step(
                     cwd=self.work_dir,
                     timeout=timeout or self.subprocess_timeout,
                     env=env,
+                    creationflags=no_window_flags(),
                 ),
             )
         except subprocess.TimeoutExpired:

src/multihead/claim_corroboration.py

@@ -17,6 +17,8 @@
 import subprocess
 from pathlib import Path
 
+from multihead.subprocess_utils import no_window_flags
+
 # Patterns to extract file paths from text
 _FILE_PATH_RE = re.compile(
     r'(?:^|[\s\'"`(])(/(?:mnt|home|tmp|usr|opt|var)[/\w._-]+(?:\.\w{1,10})?)',
@@ -71,6 +73,7 @@ def get_git_head_sha(repo_path: str = ".") -> str | None:
         result = subprocess.run(
             ["git", "-C", repo_path, "rev-parse", "HEAD"],
             capture_output=True, text=True, timeout=5,
+            creationflags=no_window_flags(),
         )
         if result.returncode == 0:
             return result.stdout.strip()
@@ -103,6 +106,7 @@ def check_sha_staleness(
         result = subprocess.run(
             ["git", "-C", repo_path, "diff", source_sha, "HEAD", "--", file_path],
             capture_output=True, text=True, timeout=10,
+            creationflags=no_window_flags(),
         )
         if result.returncode != 0:
             return {"stale": False, "penalty": 0.0, "reason": "git_error"}

src/multihead/diagnostics.py

@@ -7,6 +7,8 @@
 from pathlib import Path
 from typing import Any
 
+from multihead.subprocess_utils import no_window_flags
+
 
 @dataclass
 class DiagnosticResult:
@@ -115,6 +117,7 @@ def _check_ollama(self) -> DiagnosticResult:
             import subprocess
             result = subprocess.run(
                 ["ollama", "list"], capture_output=True, text=True, timeout=10,
+                creationflags=no_window_flags(),
             )
             if result.returncode == 0:
                 models = [l.split()[0] for l in result.stdout.strip().split("\n")[1:] if l.strip()]

src/multihead/extractors/base.py

@@ -83,6 +83,7 @@ async def map_generate(
         stage_name: str = "",
         batch_mode: bool = False,
         no_wait: bool = False,
+        on_chunk_progress: Callable[[int, int], None] | None = None,
     ) -> list[dict[str, Any] | Exception]:
         """Call generate for each prompt, optionally in parallel or batch.
 
@@ -209,6 +210,8 @@ def _on_progress(completed: int, total: int) -> None:
                     raise InterruptedError(
                         f"Shutdown at {idx}/{len(prompts)} in {stage_name or 'map_generate'}"
                     )
+                if on_chunk_progress:
+                    on_chunk_progress(idx, len(prompts))
                 try:
                     result = await BaseExtractor.call_generate(adapter, prompt)
                     results.append(result)

src/multihead/extractors/ci_extractor.py

@@ -17,6 +17,8 @@
 from pathlib import Path
 from typing import Any
 
+from multihead.subprocess_utils import no_window_flags
+
 logger = logging.getLogger(__name__)
 
 
@@ -38,6 +40,7 @@ def extract_ci_runs(
             ["gh", "api", f"repos/{owner}/{repo}/actions/runs",
              "--jq", f".workflow_runs[:{limit}]"],
             capture_output=True, text=True, timeout=30,
+            creationflags=no_window_flags(),
         )
         if result.returncode != 0:
             logger.error("gh api failed: %s", result.stderr)
@@ -124,6 +127,7 @@ def _extract_job_results(
             ["gh", "api", f"repos/{owner}/{repo}/actions/runs/{run_id}/jobs",
              "--jq", ".jobs"],
             capture_output=True, text=True, timeout=15,
+            creationflags=no_window_flags(),
         )
         if result.returncode != 0:
             return []

src/multihead/extractors/claim_extractor.py

@@ -78,8 +78,9 @@ async def extract(
             adapter, prompts, concurrency=concurrency,
             checkpoint_dir=kwargs.get("checkpoint_dir"),
             stage_name=kwargs.get("stage_name", ""),
-            batch_mode=kwargs.get("batch_mode", False),
+            batch_mode=kwargs.get("batch_mode", False),
             no_wait=kwargs.get("no_wait", False),
+            on_chunk_progress=kwargs.get("on_chunk_progress"),
         )
 
         import logging as _log

src/multihead/extractors/entity_extractor.py

@@ -40,8 +40,9 @@ async def extract(
             adapter, prompts, concurrency=concurrency,
             checkpoint_dir=kwargs.get("checkpoint_dir"),
             stage_name=kwargs.get("stage_name", ""),
-            batch_mode=kwargs.get("batch_mode", False),
+            batch_mode=kwargs.get("batch_mode", False),
             no_wait=kwargs.get("no_wait", False),
+            on_chunk_progress=kwargs.get("on_chunk_progress"),
         )
 
         for chunk, resp in zip(chunks, responses):

src/multihead/extractors/event_extractor.py

@@ -38,8 +38,9 @@ async def extract(
             adapter, prompts, concurrency=concurrency,
             checkpoint_dir=kwargs.get("checkpoint_dir"),
             stage_name=kwargs.get("stage_name", ""),
-            batch_mode=kwargs.get("batch_mode", False),
+            batch_mode=kwargs.get("batch_mode", False),
             no_wait=kwargs.get("no_wait", False),
+            on_chunk_progress=kwargs.get("on_chunk_progress"),
         )
 
         for chunk, resp in zip(chunks, responses):

src/multihead/extractors/test_results_extractor.py

@@ -22,6 +22,8 @@
 from pathlib import Path
 from typing import Any
 
+from multihead.subprocess_utils import no_window_flags
+
 logger = logging.getLogger(__name__)
 
 
@@ -243,7 +245,8 @@ def run_and_extract(
         args.extend(extra_args)
 
     try:
-        result = subprocess.run(args, capture_output=True, text=True, timeout=600)
+        result = subprocess.run(args, capture_output=True, text=True, timeout=600,
+                                creationflags=no_window_flags())
         claims = extract_from_junit_xml(xml_path)
 
         # Also parse stdout for any extra info

src/multihead/extractors/topic_assigner.py

@@ -53,6 +53,7 @@ async def extract(
             stage_name=kwargs.get("stage_name", ""),
             batch_mode=kwargs.get("batch_mode", False),
             no_wait=kwargs.get("no_wait", False),
+            on_chunk_progress=kwargs.get("on_chunk_progress"),
         )
 
         for idx, resp in zip(batch_indices, responses):

src/multihead/github_integration.py

@@ -12,6 +12,8 @@
 import subprocess
 from dataclasses import dataclass
 
+from multihead.subprocess_utils import no_window_flags
+
 logger = logging.getLogger(__name__)
 
 _GH_TIMEOUT = 30  # seconds
@@ -84,6 +86,7 @@ def infer_repo() -> str | None:
         result = subprocess.run(
             ["git", "remote", "get-url", "origin"],
             capture_output=True, text=True, timeout=10,
+            creationflags=no_window_flags(),
         )
         if result.returncode != 0:
             return None
@@ -120,6 +123,7 @@ def _run_gh(args: list[str], timeout: int = _GH_TIMEOUT) -> subprocess.Completed
     try:
         result = subprocess.run(
             cmd, capture_output=True, text=True, timeout=timeout,
+            creationflags=no_window_flags(),
         )
     except FileNotFoundError:
         raise RuntimeError(

src/multihead/init_wizard/hardware.py

@@ -9,6 +9,7 @@
 import subprocess
 from pathlib import Path
 
+from multihead.subprocess_utils import no_window_flags
 from .models import AdapterStatus, HardwareProfile
 
 logger = logging.getLogger(__name__)
@@ -54,6 +55,7 @@ def detect_hardware() -> HardwareProfile:
             result = subprocess.run(
                 ["nvidia-smi", "--query-gpu=name,memory.total", "--format=csv,noheader,nounits"],
                 capture_output=True, text=True, timeout=10,
+                creationflags=no_window_flags(),
             )
             if result.returncode == 0:
                 parts = result.stdout.strip().split(", ")
@@ -75,6 +77,7 @@ def check_adapters() -> AdapterStatus:
     try:
         result = subprocess.run(
             ["ollama", "list"], capture_output=True, text=True, timeout=10,
+            creationflags=no_window_flags(),
         )
         if result.returncode == 0:
             status.ollama_available = True

src/multihead/mcp_server/_tools_decompose.py

@@ -6,6 +6,7 @@
 
 import httpx
 
+from multihead.subprocess_utils import no_window_flags
 from ._core import _get_ks, _request, logger
 
 
@@ -181,6 +182,7 @@ async def _decompose_claude_p(goal: str, context: str = "") -> str:
             lambda: subprocess.run(
                 cmd, capture_output=True, text=True,
                 cwd=work_dir, timeout=300, env=env,
+                creationflags=no_window_flags(),
             ),
         )
     except subprocess.TimeoutExpired: