autoforge-ai 0.1.18 → 0.1.20

package/README.md

@@ -4,6 +4,12 @@
 
 A long-running autonomous coding agent powered by the Claude Agent SDK. This tool can build complete applications over multiple sessions using a two-agent pattern (initializer + coding agent). Includes a React-based UI for monitoring progress in real-time.
 
+> [!WARNING]
+> **Authentication:** Anthropic's policy states that third-party developers may not offer `claude.ai` login or subscription-based rate limits for their products (including agents built on the Claude Agent SDK) unless previously approved. Using your Claude subscription with AutoForge may risk account suspension. We recommend using an API key from [console.anthropic.com](https://console.anthropic.com/) instead.
+
+> [!NOTE]
+> **This repository is no longer actively maintained.** Most agent coding tools now ship their own long-running harnesses, making this project less necessary. Feel free to fork and continue development on your own!
+
 ## Video Tutorial
 
 [![Watch the tutorial](https://img.youtube.com/vi/nKiPOxDpcJY/hqdefault.jpg)](https://youtu.be/nKiPOxDpcJY)
@@ -34,8 +40,8 @@ irm https://claude.ai/install.ps1 | iex
 
 You need one of the following:
 
-- **Claude Pro/Max Subscription** - Use `claude login` to authenticate (recommended)
-- **Anthropic API Key** - Pay-per-use from https://console.anthropic.com/
+- **Anthropic API Key** (recommended) - Pay-per-use from https://console.anthropic.com/
+- **Claude Pro/Max Subscription** - Use `claude login` to authenticate (see warning above)
 
 ---
 
@@ -101,7 +107,7 @@ This launches the React-based web UI at `http://localhost:5173` with:
 
 The start script will:
 1. Check if Claude CLI is installed
-2. Check if you're authenticated (prompt to run `claude login` if not)
+2. Check if you're authenticated (prompt to configure authentication if not)
 3. Create a Python virtual environment
 4. Install dependencies
 5. Launch the main menu
@@ -371,7 +377,7 @@ Edit `security.py` to add or remove commands from `ALLOWED_COMMANDS`.
 Install the Claude Code CLI using the instructions in the Prerequisites section.
 
 **"Not authenticated with Claude"**
-Run `claude login` to authenticate. The start script will prompt you to do this automatically.
+Set your API key via `ANTHROPIC_API_KEY` environment variable or the Settings UI. Alternatively, run `claude login` to use subscription credentials, but note that Anthropic's policy may not permit subscription-based auth for third-party agents.
 
 **"Appears to hang on first run"**
 This is normal. The initializer agent is generating detailed test cases, which takes significant time. Watch for `[Tool: ...]` output to confirm the agent is working.

package/auth.py

@@ -53,11 +53,16 @@ AUTH_ERROR_HELP_CLI = """
 
 Claude CLI requires authentication to work.
 
-To fix this, run:
+Option 1 (Recommended): Set an API key
+  export ANTHROPIC_API_KEY=your-key-here
+  Get a key at: https://console.anthropic.com/
+
+Option 2: Use subscription login
   claude login
 
-This will open a browser window to sign in.
-After logging in, try running this command again.
+  Note: Anthropic's policy may not permit using
+  subscription auth with third-party agents.
+  API key authentication is recommended.
 ==================================================
 """
 
@@ -69,11 +74,16 @@ AUTH_ERROR_HELP_SERVER = """
 
 Claude CLI requires authentication to work.
 
-To fix this, run:
+Option 1 (Recommended): Set an API key
+  export ANTHROPIC_API_KEY=your-key-here
+  Get a key at: https://console.anthropic.com/
+
+Option 2: Use subscription login
   claude login
 
-This will open a browser window to sign in.
-After logging in, try starting the agent again.
+  Note: Anthropic's policy may not permit using
+  subscription auth with third-party agents.
+  API key authentication is recommended.
 ================================================================================
 """
 

package/autonomous_agent_demo.py

@@ -76,8 +76,8 @@ Examples:
   python autonomous_agent_demo.py --project-dir my-app --testing-ratio 0
 
 Authentication:
-  Uses Claude CLI authentication (run 'claude login' if not logged in)
-  Authentication is handled by start.bat/start.sh before this runs
+  Uses Claude CLI authentication. API key (ANTHROPIC_API_KEY) is recommended.
+  Alternatively run 'claude login', but note Anthropic's policy may restrict subscription auth.
         """,
     )
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "autoforge-ai",
-  "version": "0.1.18",
+  "version": "0.1.20",
   "description": "Autonomous coding agent with web UI - build complete apps with AI",
   "license": "AGPL-3.0",
   "bin": {

package/server/routers/agent.py

@@ -17,11 +17,11 @@ from ..utils.project_helpers import get_project_path as _get_project_path
 from ..utils.validation import validate_project_name
 
 
-def _get_settings_defaults() -> tuple[bool, str, int, bool, int, int]:
+def _get_settings_defaults() -> tuple[bool, str, int, int, int]:
     """Get defaults from global settings.
 
     Returns:
-        Tuple of (yolo_mode, model, testing_agent_ratio, playwright_headless, batch_size, testing_batch_size)
+        Tuple of (yolo_mode, model, testing_agent_ratio, batch_size, testing_batch_size)
     """
     import sys
     root = Path(__file__).parent.parent.parent
@@ -40,8 +40,6 @@ def _get_settings_defaults() -> tuple[bool, str, int, bool, int, int]:
     except (ValueError, TypeError):
         testing_agent_ratio = 1
 
-    playwright_headless = (settings.get("playwright_headless") or "true").lower() == "true"
-
     try:
         batch_size = int(settings.get("batch_size", "3"))
     except (ValueError, TypeError):
@@ -52,7 +50,7 @@ def _get_settings_defaults() -> tuple[bool, str, int, bool, int, int]:
     except (ValueError, TypeError):
         testing_batch_size = 3
 
-    return yolo_mode, model, testing_agent_ratio, playwright_headless, batch_size, testing_batch_size
+    return yolo_mode, model, testing_agent_ratio, batch_size, testing_batch_size
 
 
 router = APIRouter(prefix="/api/projects/{project_name}/agent", tags=["agent"])
@@ -101,7 +99,7 @@ async def start_agent(
     manager = get_project_manager(project_name)
 
     # Get defaults from global settings if not provided in request
-    default_yolo, default_model, default_testing_ratio, playwright_headless, default_batch_size, default_testing_batch_size = _get_settings_defaults()
+    default_yolo, default_model, default_testing_ratio, default_batch_size, default_testing_batch_size = _get_settings_defaults()
 
     yolo_mode = request.yolo_mode if request.yolo_mode is not None else default_yolo
     model = request.model if request.model else default_model
@@ -111,12 +109,13 @@ async def start_agent(
     batch_size = default_batch_size
     testing_batch_size = default_testing_batch_size
 
+    # Always run headless - the embedded browser view panel replaces desktop windows
     success, message = await manager.start(
         yolo_mode=yolo_mode,
         model=model,
         max_concurrency=max_concurrency,
         testing_agent_ratio=testing_agent_ratio,
-        playwright_headless=playwright_headless,
+        playwright_headless=True,
         batch_size=batch_size,
         testing_batch_size=testing_batch_size,
     )

package/server/routers/settings.py

@@ -111,7 +111,7 @@ async def get_settings():
         glm_mode=glm_mode,
         ollama_mode=ollama_mode,
         testing_agent_ratio=_parse_int(all_settings.get("testing_agent_ratio"), 1),
-        playwright_headless=_parse_bool(all_settings.get("playwright_headless"), default=True),
+        playwright_headless=True,  # Always headless - embedded browser view replaces desktop windows
         batch_size=_parse_int(all_settings.get("batch_size"), 3),
         testing_batch_size=_parse_int(all_settings.get("testing_batch_size"), 3),
         api_provider=api_provider,
@@ -133,8 +133,8 @@ async def update_settings(update: SettingsUpdate):
     if update.testing_agent_ratio is not None:
         set_setting("testing_agent_ratio", str(update.testing_agent_ratio))
 
-    if update.playwright_headless is not None:
-        set_setting("playwright_headless", "true" if update.playwright_headless else "false")
+    # playwright_headless is no longer user-configurable; always headless
+    # with embedded browser view panel in the UI
 
     if update.batch_size is not None:
         set_setting("batch_size", str(update.batch_size))
@@ -179,7 +179,7 @@ async def update_settings(update: SettingsUpdate):
         glm_mode=glm_mode,
         ollama_mode=ollama_mode,
         testing_agent_ratio=_parse_int(all_settings.get("testing_agent_ratio"), 1),
-        playwright_headless=_parse_bool(all_settings.get("playwright_headless"), default=True),
+        playwright_headless=True,  # Always headless - embedded browser view replaces desktop windows
         batch_size=_parse_int(all_settings.get("batch_size"), 3),
         testing_batch_size=_parse_int(all_settings.get("testing_batch_size"), 3),
         api_provider=api_provider,

package/server/services/browser_view_service.py

@@ -0,0 +1,280 @@
+"""
+Browser View Service
+====================
+
+Captures periodic screenshots from active playwright-cli browser sessions
+and streams them to the UI via WebSocket callbacks.
+
+Each agent gets an isolated browser session (e.g., coding-5, testing-0).
+This service polls those sessions with `playwright-cli screenshot` and
+delivers the frames to subscribed UI clients.
+"""
+
+import asyncio
+import base64
+import logging
+import shutil
+import threading
+from dataclasses import dataclass
+from datetime import datetime
+from pathlib import Path
+from typing import Awaitable, Callable
+
+logger = logging.getLogger(__name__)
+
+POLL_INTERVAL = 2.0  # seconds between screenshot captures
+BACKOFF_INTERVAL = 10.0  # seconds after repeated failures
+MAX_FAILURES_BEFORE_BACKOFF = 10
+MAX_FAILURES_BEFORE_STOP = 90  # ~3 minutes at normal rate before giving up
+SCREENSHOT_TIMEOUT = 5  # seconds
+
+
+@dataclass
+class SessionInfo:
+    """Metadata for an active browser session."""
+    session_name: str
+    agent_index: int
+    agent_type: str  # "coding" or "testing"
+    feature_id: int
+    feature_name: str
+    consecutive_failures: int = 0
+    stopped: bool = False
+
+
+@dataclass
+class ScreenshotData:
+    """A captured screenshot ready for delivery."""
+    session_name: str
+    agent_index: int
+    agent_type: str
+    feature_id: int
+    feature_name: str
+    image_base64: str  # base64-encoded PNG
+    timestamp: str
+
+
+class BrowserViewService:
+    """Manages screenshot capture for active agent browser sessions.
+
+    Follows the same singleton-per-project pattern as DevServerProcessManager.
+    """
+
+    def __init__(self, project_name: str, project_dir: Path):
+        self.project_name = project_name
+        self.project_dir = project_dir
+        self._active_sessions: dict[str, SessionInfo] = {}
+        self._subscribers = 0
+        self._poll_task: asyncio.Task | None = None
+        self._screenshot_callbacks: set[Callable[[ScreenshotData], Awaitable[None]]] = set()
+        self._lock = asyncio.Lock()
+        self._playwright_cli: str | None = None
+
+    def _get_playwright_cli(self) -> str | None:
+        """Find playwright-cli executable."""
+        if self._playwright_cli is not None:
+            return self._playwright_cli
+        path = shutil.which("playwright-cli")
+        if path:
+            self._playwright_cli = path
+        else:
+            logger.warning("playwright-cli not found in PATH; browser view disabled")
+        return self._playwright_cli
+
+    async def register_session(
+        self,
+        session_name: str,
+        agent_index: int,
+        agent_type: str,
+        feature_id: int,
+        feature_name: str,
+    ) -> None:
+        """Register an agent's browser session for screenshot capture."""
+        async with self._lock:
+            self._active_sessions[session_name] = SessionInfo(
+                session_name=session_name,
+                agent_index=agent_index,
+                agent_type=agent_type,
+                feature_id=feature_id,
+                feature_name=feature_name,
+            )
+            logger.debug("Registered browser session: %s", session_name)
+
+    async def unregister_session(self, session_name: str) -> None:
+        """Unregister a browser session when agent completes."""
+        async with self._lock:
+            removed = self._active_sessions.pop(session_name, None)
+            if removed:
+                logger.debug("Unregistered browser session: %s", session_name)
+                # Clean up screenshot file
+                self._cleanup_screenshot_file(session_name)
+
+    def add_screenshot_callback(self, callback: Callable[[ScreenshotData], Awaitable[None]]) -> None:
+        self._screenshot_callbacks.add(callback)
+
+    def remove_screenshot_callback(self, callback: Callable[[ScreenshotData], Awaitable[None]]) -> None:
+        self._screenshot_callbacks.discard(callback)
+
+    async def add_subscriber(self) -> None:
+        """Called when a UI client wants browser screenshots."""
+        async with self._lock:
+            self._subscribers += 1
+            if self._subscribers == 1:
+                self._start_polling()
+
+    async def remove_subscriber(self) -> None:
+        """Called when a UI client stops wanting screenshots."""
+        async with self._lock:
+            self._subscribers = max(0, self._subscribers - 1)
+            if self._subscribers == 0:
+                self._stop_polling()
+
+    async def stop(self) -> None:
+        """Clean up all sessions and stop polling."""
+        async with self._lock:
+            for session_name in list(self._active_sessions):
+                self._cleanup_screenshot_file(session_name)
+            self._active_sessions.clear()
+            self._stop_polling()
+
+    def _start_polling(self) -> None:
+        """Start the screenshot polling loop."""
+        if self._poll_task is not None and not self._poll_task.done():
+            return
+        self._poll_task = asyncio.create_task(self._poll_loop())
+        logger.info("Started browser screenshot polling for %s", self.project_name)
+
+    def _stop_polling(self) -> None:
+        """Stop the screenshot polling loop."""
+        if self._poll_task is not None and not self._poll_task.done():
+            self._poll_task.cancel()
+            self._poll_task = None
+            logger.info("Stopped browser screenshot polling for %s", self.project_name)
+
+    async def _poll_loop(self) -> None:
+        """Main polling loop - capture screenshots for all active sessions."""
+        try:
+            while True:
+                async with self._lock:
+                    sessions = list(self._active_sessions.values())
+
+                if sessions and self._screenshot_callbacks:
+                    # Capture screenshots with limited concurrency
+                    sem = asyncio.Semaphore(3)
+
+                    async def capture_with_sem(session: SessionInfo) -> None:
+                        async with sem:
+                            await self._capture_and_deliver(session)
+
+                    await asyncio.gather(
+                        *(capture_with_sem(s) for s in sessions if not s.stopped),
+                        return_exceptions=True,
+                    )
+
+                await asyncio.sleep(POLL_INTERVAL)
+        except asyncio.CancelledError:
+            pass
+        except Exception:
+            logger.warning("Browser screenshot polling crashed", exc_info=True)
+
+    async def _capture_and_deliver(self, session: SessionInfo) -> None:
+        """Capture a screenshot for a session and deliver to callbacks."""
+        cli = self._get_playwright_cli()
+        if not cli:
+            return
+
+        # Determine interval based on failure count
+        if session.consecutive_failures >= MAX_FAILURES_BEFORE_BACKOFF:
+            # In backoff mode - only capture every BACKOFF_INTERVAL/POLL_INTERVAL polls
+            # We achieve this by checking a simple modulo on failure count
+            if session.consecutive_failures % int(BACKOFF_INTERVAL / POLL_INTERVAL) != 0:
+                return
+
+        screenshot_dir = self.project_dir / ".playwright-cli"
+        screenshot_dir.mkdir(parents=True, exist_ok=True)
+        screenshot_path = screenshot_dir / f"_view_{session.session_name}.png"
+
+        try:
+            proc = await asyncio.create_subprocess_exec(
+                cli, "-s", session.session_name, "screenshot",
+                f"--filename={screenshot_path}",
+                stdout=asyncio.subprocess.PIPE,
+                stderr=asyncio.subprocess.PIPE,
+                cwd=str(self.project_dir),
+            )
+            _, stderr = await asyncio.wait_for(proc.communicate(), timeout=SCREENSHOT_TIMEOUT)
+
+            if proc.returncode != 0:
+                session.consecutive_failures += 1
+                if session.consecutive_failures >= MAX_FAILURES_BEFORE_STOP:
+                    session.stopped = True
+                    logger.debug(
+                        "Stopped polling session %s after %d failures",
+                        session.session_name, session.consecutive_failures,
+                    )
+                return
+
+            # Read and encode the screenshot
+            if not screenshot_path.exists():
+                session.consecutive_failures += 1
+                return
+
+            image_bytes = screenshot_path.read_bytes()
+            image_base64 = base64.b64encode(image_bytes).decode("ascii")
+
+            # Reset failure counter on success
+            session.consecutive_failures = 0
+            # Re-enable if previously stopped
+            session.stopped = False
+
+            screenshot = ScreenshotData(
+                session_name=session.session_name,
+                agent_index=session.agent_index,
+                agent_type=session.agent_type,
+                feature_id=session.feature_id,
+                feature_name=session.feature_name,
+                image_base64=image_base64,
+                timestamp=datetime.now().isoformat(),
+            )
+
+            # Deliver to all callbacks
+            for callback in list(self._screenshot_callbacks):
+                try:
+                    await callback(screenshot)
+                except Exception:
+                    pass  # Connection may be closed
+
+        except asyncio.TimeoutError:
+            session.consecutive_failures += 1
+        except Exception:
+            session.consecutive_failures += 1
+        finally:
+            # Clean up the screenshot file
+            try:
+                screenshot_path.unlink(missing_ok=True)
+            except Exception:
+                pass
+
+    def _cleanup_screenshot_file(self, session_name: str) -> None:
+        """Remove a session's screenshot file."""
+        try:
+            path = self.project_dir / ".playwright-cli" / f"_view_{session_name}.png"
+            path.unlink(missing_ok=True)
+        except Exception:
+            pass
+
+
+# ---------------------------------------------------------------------------
+# Global instance management (thread-safe)
+# ---------------------------------------------------------------------------
+
+_services: dict[tuple[str, str], BrowserViewService] = {}
+_services_lock = threading.Lock()
+
+
+def get_browser_view_service(project_name: str, project_dir: Path) -> BrowserViewService:
+    """Get or create a BrowserViewService for a project (thread-safe)."""
+    with _services_lock:
+        key = (project_name, str(project_dir.resolve()))
+        if key not in _services:
+            _services[key] = BrowserViewService(project_name, project_dir)
+        return _services[key]

package/server/websocket.py

@@ -16,6 +16,7 @@ from typing import Set
 from fastapi import WebSocket, WebSocketDisconnect
 
 from .schemas import AGENT_MASCOTS
+from .services.browser_view_service import get_browser_view_service
 from .services.chat_constants import ROOT_DIR
 from .services.dev_server_manager import get_devserver_manager
 from .services.process_manager import get_manager
@@ -787,8 +788,39 @@ async def project_websocket(websocket: WebSocket, project_name: str):
     # Create orchestrator tracker for observability
     orchestrator_tracker = OrchestratorTracker()
 
+    # Get browser view service for embedded browser screenshots
+    browser_view_service = get_browser_view_service(project_name, project_dir)
+    browser_view_subscribed = False
+    # Counter to mirror orchestrator's testing session naming (testing-0, testing-1, ...)
+    testing_session_counter = 0
+    # Deferred session registration: store metadata at agent start, register on first browser command.
+    # This avoids premature polling failures when agents spend time reading/planning before opening a browser.
+    # Key: session_name -> registration kwargs
+    pending_browser_sessions: dict[str, dict] = {}
+    # Track which feature IDs map to which session names (for deferred lookup)
+    feature_to_session: dict[int, str] = {}
+
+    async def on_screenshot(screenshot):
+        """Handle browser screenshot - send to this WebSocket."""
+        try:
+            await websocket.send_json({
+                "type": "browser_screenshot",
+                "sessionName": screenshot.session_name,
+                "agentIndex": screenshot.agent_index,
+                "agentType": screenshot.agent_type,
+                "featureId": screenshot.feature_id,
+                "featureName": screenshot.feature_name,
+                "imageData": screenshot.image_base64,
+                "timestamp": screenshot.timestamp,
+            })
+        except Exception:
+            pass  # Connection may be closed
+
+    browser_view_service.add_screenshot_callback(on_screenshot)
+
     async def on_output(line: str):
         """Handle agent output - broadcast to this WebSocket."""
+        nonlocal testing_session_counter
         try:
             # Extract feature ID from line if present
             feature_id = None
@@ -817,6 +849,48 @@ async def project_websocket(websocket: WebSocket, project_name: str):
             if agent_update:
                 await websocket.send_json(agent_update)
 
+                # Register/unregister browser sessions based on agent lifecycle
+                update_state = agent_update.get("state")
+                update_type = agent_update.get("agentType", "coding")
+                update_feature_id = agent_update.get("featureId", 0)
+                update_feature_name = agent_update.get("featureName", "")
+                update_agent_index = agent_update.get("agentIndex", 0)
+
+                if update_state == "thinking" and agent_update.get("thought") in ("Starting work...", "Starting batch work..."):
+                    # Agent just started - defer browser session registration until
+                    # we detect an actual playwright-cli open/goto command.  This avoids
+                    # polling failures while the agent is still reading code / planning.
+                    if update_type == "coding":
+                        session_name = f"coding-{update_feature_id}"
+                    else:
+                        session_name = f"testing-{testing_session_counter}"
+                        testing_session_counter += 1
+                    pending_browser_sessions[session_name] = dict(
+                        session_name=session_name,
+                        agent_index=update_agent_index,
+                        agent_type=update_type,
+                        feature_id=update_feature_id,
+                        feature_name=update_feature_name,
+                    )
+                    feature_to_session[update_feature_id] = session_name
+                elif update_state in ("success", "error"):
+                    # Agent completed - unregister browser session
+                    if update_type == "coding":
+                        session_name = f"coding-{update_feature_id}"
+                        await browser_view_service.unregister_session(session_name)
+                        pending_browser_sessions.pop(session_name, None)
+                        feature_to_session.pop(update_feature_id, None)
+                    # Testing sessions are cleaned up on orchestrator stop
+
+            # Detect playwright-cli browser commands and activate deferred sessions
+            if feature_id is not None and "playwright-cli" in line and any(
+                kw in line for kw in ("open ", "goto ", "open\t", "goto\t")
+            ):
+                sess_name = feature_to_session.get(feature_id)
+                if sess_name and sess_name in pending_browser_sessions:
+                    reg = pending_browser_sessions.pop(sess_name)
+                    await browser_view_service.register_session(**reg)
+
             # Also check for orchestrator events and emit orchestrator_update messages
             orch_update = await orchestrator_tracker.process_line(line)
             if orch_update:
@@ -826,6 +900,7 @@ async def project_websocket(websocket: WebSocket, project_name: str):
 
     async def on_status_change(status: str):
         """Handle status change - broadcast to this WebSocket."""
+        nonlocal testing_session_counter
         try:
             await websocket.send_json({
                 "type": "agent_status",
@@ -835,6 +910,10 @@ async def project_websocket(websocket: WebSocket, project_name: str):
             if status in ("stopped", "crashed"):
                 await agent_tracker.reset()
                 await orchestrator_tracker.reset()
+                await browser_view_service.stop()
+                testing_session_counter = 0
+                pending_browser_sessions.clear()
+                feature_to_session.clear()
         except Exception:
             pass  # Connection may be closed
 
@@ -908,10 +987,23 @@ async def project_websocket(websocket: WebSocket, project_name: str):
                 data = await websocket.receive_text()
                 message = json.loads(data)
 
+                msg_type = message.get("type")
+
                 # Handle ping
-                if message.get("type") == "ping":
+                if msg_type == "ping":
                     await websocket.send_json({"type": "pong"})
 
+                # Handle browser view subscribe/unsubscribe
+                elif msg_type == "browser_view_subscribe":
+                    if not browser_view_subscribed:
+                        browser_view_subscribed = True
+                        await browser_view_service.add_subscriber()
+
+                elif msg_type == "browser_view_unsubscribe":
+                    if browser_view_subscribed:
+                        browser_view_subscribed = False
+                        await browser_view_service.remove_subscriber()
+
             except WebSocketDisconnect:
                 break
             except json.JSONDecodeError:
@@ -935,5 +1027,10 @@ async def project_websocket(websocket: WebSocket, project_name: str):
         devserver_manager.remove_output_callback(on_dev_output)
         devserver_manager.remove_status_callback(on_dev_status_change)
 
+        # Unregister browser view callbacks and subscriber
+        browser_view_service.remove_screenshot_callback(on_screenshot)
+        if browser_view_subscribed:
+            await browser_view_service.remove_subscriber()
+
         # Disconnect from manager
         await manager.disconnect(websocket, project_name)

package/start.py

@@ -255,7 +255,7 @@ def run_spec_creation(project_dir: Path) -> bool:
             print(f"Please ensure app_spec.txt exists in: {get_project_prompts_dir(project_dir)}")
             # If failed with non-zero exit and no spec, might be auth issue
             if result.returncode != 0:
-                print("\nIf you're having authentication issues, try running: claude login")
+                print("\nIf you're having authentication issues, set ANTHROPIC_API_KEY or try: claude login")
             return False
 
     except FileNotFoundError:
@@ -416,7 +416,7 @@ def run_agent(project_name: str, project_dir: Path) -> None:
                 print(f"\nAgent error:\n{stderr_output.strip()}")
                 # Still hint about auth if exit was unexpected
                 if "error" in stderr_output.lower() or "exception" in stderr_output.lower():
-                    print("\nIf this is an authentication issue, try running: claude login")
+                    print("\nIf this is an authentication issue, set ANTHROPIC_API_KEY or try: claude login")
 
     except KeyboardInterrupt:
         print("\n\nAgent interrupted. Run again to resume.")