alpiecode 5.1.1__tar.gz → 6.1.0__tar.gz

{alpiecode-5.1.1 → alpiecode-6.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: alpiecode
-Version: 5.1.1
+Version: 6.1.0
 Summary: AlpieCode — Autonomous AI Coding Agent CLI powered by local 169Pi GGUF VLM
 Requires-Python: >=3.9
 Requires-Dist: huggingface_hub>=0.20.0

{alpiecode-5.1.1 → alpiecode-6.1.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "alpiecode"
-version = "5.1.1"
+version = "6.1.0"
 description = "AlpieCode — Autonomous AI Coding Agent CLI powered by local 169Pi GGUF VLM"
 requires-python = ">=3.9"
 dependencies = [

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/alpiecode.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: alpiecode
-Version: 5.1.1
+Version: 6.1.0
 Summary: AlpieCode — Autonomous AI Coding Agent CLI powered by local 169Pi GGUF VLM
 Requires-Python: >=3.9
 Requires-Dist: huggingface_hub>=0.20.0

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/alpiecode.egg-info/SOURCES.txt

@@ -25,6 +25,7 @@ src/codeagent/local_model.py
 src/codeagent/media.py
 src/codeagent/memory.py
 src/codeagent/orchestrator.py
+src/codeagent/progress.py
 src/codeagent/prompt.py
 src/codeagent/server.py
 src/codeagent/session.py

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/__init__.py

@@ -16,7 +16,7 @@ Supports:
 from pathlib import Path
 from typing import Optional, Any, Dict
 
-__version__ = "2.0.4"
+__version__ = "6.1.0"
 
 
 def run(task: str, workdir: str = ".", reasoning_level: str = "medium", **kwargs):

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/agent.py

@@ -37,6 +37,7 @@ _build_system_prompt = lambda workdir, is_offline=False: PromptBuilder().build_s
 try:
     from rich.console import Console
     from rich.markdown import Markdown
+    from rich.markup import escape
     from rich.panel import Panel
     from rich.rule import Rule
     from rich.text import Text
@@ -46,6 +47,9 @@ try:
 except ImportError:
     HAS_RICH = False
 
+    def escape(text: str) -> str:
+        return text
+
     class _FallbackConsole:
         def print(self, *args, **kwargs):
             kwargs.pop("style", None)
@@ -78,7 +82,7 @@ def _print_tool_call(turn: int, name: str, args: dict):
     if HAS_RICH:
         args_str = json.dumps(display_args, indent=2)
         console.print(f"\n🔧 [bold cyan]Tool:[/bold cyan] [bold]{name}[/bold]", highlight=False)
-        console.print(f"   {args_str}", style="cyan", highlight=False)
+        console.print(Text(f"   {args_str}", style="cyan"))
     else:
         console.print(f"\n🔧 Tool: {name}({display_args})")
 
@@ -86,7 +90,7 @@ def _print_tool_call(turn: int, name: str, args: dict):
 def _print_tool_result(result: str):
     truncated = result[:1500] + ("..." if len(result) > 1500 else "")
     if HAS_RICH:
-        console.print(f"   → {truncated}", style="green", highlight=False)
+        console.print(Text(f"   → {truncated}", style="green"))
     else:
         console.print(f"   → {truncated}")
 
@@ -97,7 +101,7 @@ def _print_assistant_message(content: str):
             md = Markdown(content)
             console.print(Panel(md, title="🤖 Assistant", border_style="green", padding=(0, 1)))
         except Exception:
-            console.print(Panel(content, title="🤖 Assistant", border_style="green", padding=(0, 1)))
+            console.print(Panel(Text(content), title="🤖 Assistant", border_style="green", padding=(0, 1)))
     else:
         console.print(f"\n🤖 Assistant: {content}")
 
@@ -225,7 +229,7 @@ def run_agent(
             data = event.data
             if HAS_RICH:
                 console.rule("[bold blue]Agent Started[/bold blue]")
-                console.print(f"📋 Task: {task.splitlines()[0]}", style="bold")
+                console.print(Text(f"📋 Task: {task.splitlines()[0]}", style="bold"))
                 if github_repo:
                     console.print(f"🐙 GitHub Repo: {github_repo}", style="cyan")
                 if image_path:
@@ -288,8 +292,9 @@ def run_agent(
             _checkpoint(workdir, "checkpoint: response")
 
         elif event.type == "fallback" and verbose:
+            err_str = escape(str(event.data.get('error', '')))
             if HAS_RICH:
-                console.print(f"\n⚠️  [bold yellow]Online Server Error / Timeout[/bold yellow] ({event.data['error']})", style="yellow")
+                console.print(f"\n⚠️  [bold yellow]Online Server Error / Timeout[/bold yellow] ({err_str})", style="yellow")
                 console.print("🔄 [bold cyan]Auto-falling back to local GGUF engine...[/bold cyan]", style="cyan")
             else:
                 print(f"\n⚠️ Online Server Error: {event.data['error']}")
@@ -297,10 +302,41 @@ def run_agent(
 
         elif event.type == "error" and verbose:
             if HAS_RICH:
-                console.print(f"\n❌ [bold red]Model Error[/bold red]\n   Error: {event.data['error']}\n")
+                console.print(Text(f"\n❌ Model Error\n   Error: {event.data['error']}\n", style="bold red"))
             else:
                 print(f"\n❌ Model Error: {event.data['error']}")
 
+        elif event.type == "stall_intervention" and verbose:
+            if HAS_RICH:
+                console.print(
+                    f"🔄 [bold yellow]Progress stall detected[/bold yellow] "
+                    f"(turn {event.data['turn']}, {event.data['consecutive_stalls']} stalled turns, "
+                    f"intervention #{event.data['interventions']})",
+                    style="yellow"
+                )
+            else:
+                print(f"🔄 Progress stall detected (turn {event.data['turn']})")
+
+        elif event.type == "turn_progress" and debug:
+            snap = event.data
+            status = "✅" if snap["had_progress"] else "⚠️"
+            if HAS_RICH:
+                console.print(
+                    Text(f"  {status} Progress: created={snap['files_created']}, "
+                    f"modified={snap['files_modified']}, stalls={snap['consecutive_stalls']}"),
+                    style="dim"
+                )
+
+        elif event.type == "safety_ceiling" and verbose:
+            if HAS_RICH:
+                console.print(
+                    f"\n🛑 [bold red]Safety ceiling ({event.data['ceiling']} turns) reached.[/bold red]\n"
+                    "This is an emergency stop — the agent may be stuck in an unrecoverable loop.",
+                    style="bold red"
+                )
+            else:
+                print(f"\n🛑 Safety ceiling ({event.data['ceiling']}) reached.")
+
         elif event.type == "done":
             summary = event.data["summary"]
             _checkpoint(workdir, "checkpoint: done")
@@ -311,7 +347,18 @@ def run_agent(
                     console.rule("[bold yellow]💬 Agent Replied[/bold yellow]")
 
         elif event.type == "max_turns_reached" and verbose:
-            console.print(f"\n⚠️  Max turns ({event.data['max_turns']}) reached without completion.", style="bold yellow")
+            progress = event.data.get("progress", {})
+            if HAS_RICH:
+                console.print(f"\n⚠️  [bold yellow]Safety ceiling ({event.data['max_turns']}) reached.[/bold yellow]", style="bold yellow")
+                if progress:
+                    console.print(
+                        f"   Progress: {progress.get('progress_turns', 0)}/{progress.get('total_turns', 0)} turns made progress, "
+                        f"{progress.get('files_created', 0)} files created, "
+                        f"{progress.get('stall_interventions', 0)} stall interventions",
+                        style="dim"
+                    )
+            else:
+                print(f"\n⚠️ Safety ceiling ({event.data['max_turns']}) reached.")
 
     return session.context.messages if "session" in locals() else []
 
@@ -385,7 +432,10 @@ def run_chat(workdir: Path, cfg: Config, verbose: bool = True) -> None:
                 _checkpoint(workdir, "checkpoint: done")
 
             elif event.type == "error":
-                console.print(f"❌ Model error: {event.data['error']}", style="bold red" if HAS_RICH else None)
+                if HAS_RICH:
+                    console.print(Text(f"❌ Model error: {event.data['error']}", style="bold red"))
+                else:
+                    print(f"❌ Model error: {event.data['error']}")
 
             elif event.type == "done":
                 break

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/compaction.py

@@ -20,7 +20,7 @@ MAX_CONTEXT_TOKENS = 262_144
 # Start compacting when we hit this percentage of the context window
 COMPACT_THRESHOLD = 0.70
 # Number of recent turns to always keep intact
-KEEP_RECENT_TURNS = 20
+KEEP_RECENT_TURNS = 12
 # Approximate chars per token (rough heuristic)
 CHARS_PER_TOKEN = 4
 

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/config.py

@@ -26,7 +26,7 @@ DEFAULTS = {
     "model_repo": "169Pi/Alpie_learn_prototype_GGUF_NEW",
     "api_key": "not-needed",
     "hf_token": None,
-    "max_turns": 20,
+    "max_turns": 200,
     "temperature": 0.1,
     "max_tokens": 8192,
     "enable_thinking": False,  # Reasoning OFF by default
@@ -88,7 +88,7 @@ class Config:
     model_repo: str = "169Pi/Alpie_learn_prototype_GGUF_NEW"  # HuggingFace repo for offline GGUF
     api_key: str = "not-needed"
     hf_token: Optional[str] = None
-    max_turns: int = 20
+    max_turns: int = 200  # Safety ceiling only — agent runs until DONE
     temperature: float = 0.1
     max_tokens: int = 8192
     enable_thinking: bool = False
@@ -117,7 +117,7 @@ def load_config() -> Config:
                 data["enable_thinking"] = False
                 data["temperature"] = 0.1
                 data["max_tokens"] = 8192
-                data["max_turns"] = 20
+                data["max_turns"] = 200
                 data["config_version"] = CONFIG_VERSION
                 needs_save = True
         except Exception:

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/discovery.py

@@ -49,7 +49,6 @@ class TaskContext:
     dependencies: List[str] = field(default_factory=list)
 
     # Budget (computed from complexity)
-    max_turns: int = 10
     max_tokens: int = 8192
     tool_set: str = "core"          # none, core, full
     enable_thinking: bool = False
@@ -178,16 +177,18 @@ def detect_environment() -> dict:
 
 
 def _detect_shell(os_name: str) -> str:
-    """Detect available shell, preferring WSL on Windows like Claude/Codex agents."""
+    """Detect available shell, aligned with tools.py execution path."""
 
     if os_name == "windows":
-        # On Windows: prefer WSL (like Claude, Codex, Gemini agents do)
+        # Check if bash binary exists (Git Bash / MSYS)
+        if shutil.which("bash"):
+            return "bash"
         if _is_wsl_available():
             return "wsl"
         # Fallback to PowerShell
         if shutil.which("powershell") or shutil.which("pwsh"):
             return "powershell"
-        return "cmd"
+        return "cmd" 
 
     if os_name in ("linux", "wsl"):
         return "bash"  # Standard on Linux/WSL
@@ -466,13 +467,14 @@ def compute_complexity(intent: str, repo_info: dict, task: str) -> str:
 
     # -- MEDIUM: multi-step tasks --
     medium_keywords = [
-        "api", "rest api", "graphql", "server",
-        "game", "snake", "tetris", "chess", "pong", "sudoku",
-        "website", "web page", "web app", "webapp",
-        "test suite", "unit tests", "integration test",
-        "dashboard", "portfolio",
-        "react", "vue", "angular", "next.js",
-        "django", "flask app", "fastapi app",
+        "api", "rest api", "graphql", "server", "backend", "frontend",
+        "game", "snake", "tetris", "chess", "pong", "sudoku", "flappy", "arcade",
+        "website", "web page", "web app", "webapp", "html", "css", "javascript",
+        "test suite", "unit tests", "integration test", "e2e test",
+        "dashboard", "portfolio", "application", "app with", "notes app", "todo app",
+        "react", "vue", "angular", "next.js", "tailwind",
+        "django", "flask app", "fastapi app", "express",
+        "refactor", "migrate", "redesign", "restructure", "investigate",
     ]
     if any(kw in task_lower for kw in medium_keywords):
         return "medium"
@@ -497,25 +499,21 @@ def compute_complexity(intent: str, repo_info: dict, task: str) -> str:
 
 COMPLEXITY_CONFIG = {
     "qa": {
-        "max_turns": 3,
         "max_tokens": 4096,
         "tool_set": "none",
         "enable_thinking": False,
     },
     "low": {
-        "max_turns": 10,
         "max_tokens": 8192,
         "tool_set": "core",
         "enable_thinking": False,
     },
     "medium": {
-        "max_turns": 20,
-        "max_tokens": 8192,
+        "max_tokens": 16384,
         "tool_set": "full",
-        "enable_thinking": False,
+        "enable_thinking": True,
     },
     "high": {
-        "max_turns": 40,
         "max_tokens": 16384,
         "tool_set": "full",
         "enable_thinking": True,
@@ -565,7 +563,6 @@ def build_task_context(task: str, workdir: Path) -> TaskContext:
         entry_points=repo["entry_points"],
         dependencies=repo["dependencies"],
         # Budget
-        max_turns=budget["max_turns"],
         max_tokens=budget["max_tokens"],
         tool_set=budget["tool_set"],
         enable_thinking=budget["enable_thinking"],

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/guardian.py

@@ -16,12 +16,17 @@ from typing import Tuple
 try:
     from rich.console import Console
     from rich.panel import Panel
+    from rich.text import Text
+    from rich.markup import escape
     console = Console()
     HAS_RICH = True
 except ImportError:
     HAS_RICH = False
     console = None
 
+    def escape(text: str) -> str:
+        return text
+
 
 class RiskLevel(Enum):
     SAFE = "safe"
@@ -134,19 +139,19 @@ def gate_command(command: str, auto_approve: bool = False) -> bool:
     if risk == RiskLevel.WARNING:
         if auto_approve:
             if HAS_RICH:
-                console.print(f"   ⚠️  [yellow]{reason}[/yellow]", highlight=False)
+                console.print(Text(f"   ⚠️  {reason}", style="yellow"))
             return True
         # In interactive mode, show warning but proceed
         if HAS_RICH:
-            console.print(f"   ⚠️  [yellow]{reason}[/yellow]", highlight=False)
+            console.print(Text(f"   ⚠️  {reason}", style="yellow"))
         return True
 
     if risk == RiskLevel.DANGEROUS:
         if HAS_RICH:
             console.print(Panel(
                 f"[bold red]🛑 BLOCKED — Dangerous Command[/bold red]\n\n"
-                f"Command: [cyan]{command}[/cyan]\n"
-                f"Reason: {reason}\n\n"
+                f"Command: [cyan]{escape(command)}[/cyan]\n"
+                f"Reason: {escape(reason)}\n\n"
                 f"This command has been blocked for safety.\n"
                 f"If you need to run it, do so manually in your terminal.",
                 border_style="red",

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/orchestrator.py

@@ -15,6 +15,7 @@ from .cache import get_cache
 from .config import Config, is_server_reachable
 from .memory import extract_and_save_memories
 from .discovery import build_task_context, COMPLEXITY_CONFIG
+from .progress import ProgressMonitor
 from .prompt import PromptBuilder, classify_task
 from .session import Session, SessionManager
 
@@ -72,13 +73,11 @@ class AgentOrchestrator:
 
         comp_cfg = COMPLEXITY_CONFIG.get(complexity, COMPLEXITY_CONFIG["low"])
 
-        # ── Determine effective max_turns and max_tokens ──
-        effective_max_turns = task_context.max_turns
+        # ── Determine effective max_tokens ──
         effective_max_tokens = task_context.max_tokens
 
-        # User override: if they set --max-turns explicitly, respect it
-        if cfg.max_turns != 20:  # 20 is new default, so non-default = explicit
-            effective_max_turns = cfg.max_turns
+        # Safety ceiling: hard emergency brake (should never be hit naturally)
+        safety_ceiling = cfg.max_turns if cfg.max_turns != 200 else 200
 
         # ── Response cache check ──
         is_cacheable = not any([image_path, video_path, url, github_repo])
@@ -137,36 +136,51 @@ class AgentOrchestrator:
             "complexity": complexity,
         })
 
+
+
         # ── Adaptive thinking ──
         enable_thinking = cfg.enable_thinking or task_context.enable_thinking
         if enable_thinking and complexity in ("qa", "low"):
             enable_thinking = False
             yield AgentEvent("adaptive_mode", {"message": "Simple task detected, skipping deep reasoning."})
 
-        # ── Turn loop ──
-        wrap_up_injected = False
+        # ── Goal-driven turn loop (no fixed limit) ──
+        progress_monitor = ProgressMonitor()
+        turn = 0
+
+        while True:
+            turn += 1
 
-        for turn in range(effective_max_turns):
             if session.cancelled:
-                yield AgentEvent("cancelled", {"turn": turn + 1})
+                yield AgentEvent("cancelled", {"turn": turn})
                 break
 
             # Context compaction
             if session.context.check_and_compact():
-                yield AgentEvent("compaction", {"turn": turn + 1})
-
-            # ── Wrap-up injection at 80% of turns ──
-            if not wrap_up_injected and turn >= int(effective_max_turns * 0.8):
-                wrap_up_injected = True
-                remaining = effective_max_turns - turn
-                session.context.add_user_message(
-                    f"[SYSTEM] You have {remaining} turns remaining. "
-                    "Finish your current work now. If code is written and tested, "
-                    "output DONE: <summary>. If code has errors, make one final fix attempt."
-                )
-                yield AgentEvent("wrap_up", {"turn": turn + 1, "remaining": remaining})
+                yield AgentEvent("compaction", {"turn": turn})
+
+            # ── Stall Detection & Corrective Intervention ──
+            if progress_monitor.is_stalled(threshold=3):
+                advice = progress_monitor.get_stall_advice()
+                session.context.add_user_message(advice)
+                yield AgentEvent("stall_intervention", {
+                    "turn": turn,
+                    "consecutive_stalls": progress_monitor.consecutive_stalls,
+                    "interventions": progress_monitor.stall_interventions,
+                })
+                # After 3 interventions (= 9+ stalled turns), force wrap-up
+                if progress_monitor.stall_interventions >= 3:
+                    session.context.add_user_message(
+                        "[SYSTEM] Multiple stall interventions have not resolved the issue. "
+                        "Finish now with whatever you have. Output DONE: <summary of what was completed>."
+                    )
+
+            # ── Safety ceiling (emergency only) ──
+            if turn > safety_ceiling:
+                yield AgentEvent("safety_ceiling", {"turn": turn, "ceiling": safety_ceiling})
+                break
 
-            yield AgentEvent("turn_start", {"turn": turn + 1})
+            yield AgentEvent("turn_start", {"turn": turn})
 
             try:
                 if enable_thinking:
@@ -218,7 +232,7 @@ class AgentOrchestrator:
                     results = session.executor.execute_tool_calls(tool_calls)
                     for res in results:
                         yield AgentEvent("tool_result", {
-                            "turn": turn + 1,
+                            "turn": turn,
                             "id": res.tool_call_id,
                             "name": res.name,
                             "content": res.content,
@@ -246,7 +260,7 @@ class AgentOrchestrator:
             if tool_calls:
                 for tc in tool_calls:
                     yield AgentEvent("tool_call", {
-                        "turn": turn + 1,
+                        "turn": turn,
                         "id": tc.id,
                         "name": tc.name,
                         "arguments": tc.arguments,
@@ -256,7 +270,7 @@ class AgentOrchestrator:
 
                 for res in results:
                     yield AgentEvent("tool_result", {
-                        "turn": turn + 1,
+                        "turn": turn,
                         "id": res.tool_call_id,
                         "name": res.name,
                         "content": res.content,
@@ -264,6 +278,18 @@ class AgentOrchestrator:
                     })
                     session.context.add_tool_result(res.tool_call_id, res.content)
 
+                # ── Record progress for stall detection ──
+                tc_dicts = [{"name": tc.name, "arguments": tc.arguments} for tc in tool_calls]
+                res_dicts = [{"content": res.content, "name": res.name} for res in results]
+                snap = progress_monitor.record_turn(turn, tc_dicts, res_dicts)
+                yield AgentEvent("turn_progress", {
+                    "turn": turn,
+                    "had_progress": snap.had_progress,
+                    "files_created": list(snap.files_created),
+                    "files_modified": list(snap.files_modified),
+                    "consecutive_stalls": progress_monitor.consecutive_stalls,
+                })
+
                 continue
 
             # Text-only response = done
@@ -285,5 +311,6 @@ class AgentOrchestrator:
             yield AgentEvent("done", {"summary": "Task completed."})
             return
 
-        yield AgentEvent("max_turns_reached", {"max_turns": effective_max_turns})
+        status = progress_monitor.get_status_summary()
+        yield AgentEvent("max_turns_reached", {"max_turns": safety_ceiling, "progress": status})
         extract_and_save_memories(session.workdir, session.context.messages)

alpiecode-6.1.0/src/codeagent/progress.py

@@ -0,0 +1,175 @@
+"""
+Progress Monitor for AlpieCode.
+
+Tracks per-turn agent progress by observing tool calls and results.
+Detects stalls (zero net progress) and thrashing (delete-after-create loops).
+Does NOT kill the agent — only signals the orchestrator to inject corrective prompts.
+"""
+
+from dataclasses import dataclass, field
+from typing import Dict, List, Set, Optional
+from pathlib import Path
+
+
+@dataclass
+class TurnSnapshot:
+    """Captures what happened in a single turn."""
+    turn: int
+    tools_called: List[str] = field(default_factory=list)
+    files_created: Set[str] = field(default_factory=set)
+    files_modified: Set[str] = field(default_factory=set)
+    files_deleted: Set[str] = field(default_factory=set)
+    bash_exit_codes: List[int] = field(default_factory=list)
+    errors: List[str] = field(default_factory=list)
+    had_progress: bool = False
+
+
+class ProgressMonitor:
+    """Tracks agent progress across turns.
+
+    Progress = files created, files modified, successful bash runs,
+    or errors changing (not repeating the same error).
+
+    Stall = 3+ consecutive turns with zero net progress.
+    Thrashing = deleting files the agent itself created.
+    """
+
+    def __init__(self):
+        self.history: List[TurnSnapshot] = []
+        self.all_files_created: Set[str] = set()
+        self.consecutive_stalls: int = 0
+        self.stall_interventions: int = 0
+
+    def record_turn(self, turn: int, tool_calls: list, tool_results: list) -> TurnSnapshot:
+        """Record what happened in a turn and determine if progress was made."""
+        snap = TurnSnapshot(turn=turn)
+
+        for tc in tool_calls:
+            name = tc.get("name", "") if isinstance(tc, dict) else getattr(tc, "name", "")
+            args = tc.get("arguments", {}) if isinstance(tc, dict) else getattr(tc, "arguments", {})
+            if not isinstance(args, dict):
+                args = {}
+            snap.tools_called.append(name)
+
+            if name == "write_file":
+                path = args.get("path", "")
+                snap.files_created.add(path)
+                self.all_files_created.add(path)
+
+            elif name == "edit_file":
+                path = args.get("path", "")
+                snap.files_modified.add(path)
+
+            elif name == "bash":
+                cmd = args.get("command", "")
+                cmd_parts = cmd.strip().split()
+                if cmd_parts and cmd_parts[0] in ("rm", "del", "Remove-Item"):
+                    for part in cmd_parts[1:]:
+                        if not part.startswith("-"):
+                            clean = Path(part.strip("\'\"")).name
+                            snap.files_deleted.add(clean)
+
+        # Parse tool results for exit codes and errors
+        for res in tool_results:
+            content = res.get("content", "") if isinstance(res, dict) else getattr(res, "content", "")
+            if '"exit_code": 0' in content or '"exit_code":0' in content:
+                snap.bash_exit_codes.append(0)
+            elif '"exit_code":' in content:
+                snap.bash_exit_codes.append(1)
+                snap.errors.append(f"command failed: {content[:100]}")
+            if content.lower().startswith("error:") or "error:" in content[:80].lower():
+                snap.errors.append(content[:100])
+
+        # Determine progress
+        snap.had_progress = bool(
+            snap.files_created
+            or snap.files_modified
+            or (snap.bash_exit_codes and 0 in snap.bash_exit_codes)
+        )
+
+        # No tools called at all = not progress (empty turn)
+        if not snap.tools_called:
+            snap.had_progress = True  # text-only response = model is finishing
+
+        # Detect thrashing: deleting a file that was created recently
+        created_names = {Path(f).name for f in self.all_files_created}
+        thrashing = snap.files_deleted & created_names
+        if thrashing:
+            snap.had_progress = False
+
+        # Detect identical consecutive errors (stall)
+        if len(self.history) >= 2 and snap.errors:
+            prev_errors = set(e[:60] for e in self.history[-1].errors)
+            curr_errors = set(e[:60] for e in snap.errors)
+            if curr_errors and curr_errors == prev_errors:
+                snap.had_progress = False
+
+        # Detect pure read-only turns with errors as non-progress
+        if snap.errors and not snap.files_created and not snap.files_modified:
+            if not (snap.bash_exit_codes and 0 in snap.bash_exit_codes):
+                snap.had_progress = False
+
+        # Update stall counter
+        if snap.had_progress:
+            self.consecutive_stalls = 0
+        else:
+            self.consecutive_stalls += 1
+
+        self.history.append(snap)
+        return snap
+
+    def is_stalled(self, threshold: int = 3) -> bool:
+        """Returns True if the agent has made zero net progress for N consecutive turns."""
+        return self.consecutive_stalls >= threshold
+
+    def get_stall_advice(self) -> str:
+        """Generate a corrective prompt for the model when stalled."""
+        self.stall_interventions += 1
+
+        recent = self.history[-3:] if len(self.history) >= 3 else self.history
+        deleted = set()
+        for snap in recent:
+            deleted.update(snap.files_deleted)
+
+        created_names = {Path(f).name for f in self.all_files_created}
+        thrashing = deleted & created_names
+
+        if thrashing:
+            return (
+                f"[SYSTEM - PROGRESS MONITOR] You have been deleting files you previously "
+                f"created ({', '.join(thrashing)}). STOP deleting and rewriting from scratch. "
+                f"Use edit_file to modify specific sections, or overwrite directly with write_file. "
+                f"Take a step back: what is the simplest path to a working solution?"
+            )
+
+        if all(snap.errors or (snap.bash_exit_codes and 0 not in snap.bash_exit_codes) for snap in recent):
+            return (
+                "[SYSTEM - PROGRESS MONITOR] You have encountered errors for 3 consecutive turns. "
+                "STOP retrying the same approach. Instead:\n"
+                "1. Use read_file to examine the FULL current state of the file(s) you\'re editing\n"
+                "2. Identify the root cause of the error (not the symptom)\n"
+                "3. Make ONE comprehensive fix that addresses all issues\n"
+                "If the task approach is fundamentally wrong, start with a simpler design."
+            )
+
+        return (
+            "[SYSTEM - PROGRESS MONITOR] No measurable progress detected for 3 turns. "
+            "You may be stuck in a loop. Either:\n"
+            "1. Complete your current work and output DONE: <summary>\n"
+            "2. Try a completely different approach to the problem\n"
+            "3. If the code is written and working, verify with bash and finish."
+        )
+
+    def get_status_summary(self) -> dict:
+        """Return a summary of overall progress for logging."""
+        total_files = len(self.all_files_created)
+        total_turns = len(self.history)
+        success_turns = sum(1 for s in self.history if s.had_progress)
+        return {
+            "total_turns": total_turns,
+            "progress_turns": success_turns,
+            "stall_turns": total_turns - success_turns,
+            "files_created": total_files,
+            "stall_interventions": self.stall_interventions,
+            "consecutive_stalls": self.consecutive_stalls,
+        }

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/prompt.py

@@ -333,25 +333,21 @@ def is_simple_task(task: str) -> bool:
 
 COMPLEXITY_CONFIG = {
     "qa": {
-        "max_turns": 3,
         "max_tokens": 4096,
         "tools": "none",  # No tools for Q&A
         "prompt": "default",
     },
     "low": {
-        "max_turns": 10,
         "max_tokens": 8192,
         "tools": "core",  # 5 core tools
         "prompt": "default",
     },
     "medium": {
-        "max_turns": 20,
-        "max_tokens": 8192,
+        "max_tokens": 16384,
         "tools": "full",  # All 15 tools
         "prompt": "default",
     },
     "high": {
-        "max_turns": 40,
         "max_tokens": 16384,
         "tools": "full",  # All 15 tools
         "prompt": "high",  # Detailed system prompt
@@ -408,7 +404,7 @@ REPO_CONTEXT_TEMPLATE = """\
 {extra}"""
 
 INTENT_CREATE = """\
-## Task Intent: Create New Code
+## Task Intent: Create New Code (Single-file / Simple Tasks)
 - Write the COMPLETE, WORKING code on Turn 1 using write_file
 - Do NOT explore the filesystem first -- start coding immediately
 - Include ALL imports, ALL functions, ALL logic -- no stubs, no TODOs
@@ -417,6 +413,23 @@ INTENT_CREATE = """\
 - When everything works: DONE: <summary>
 """
 
+INTENT_CREATE_COMPLEX = """\
+## Task Intent: Create Multi-Component Code (Medium/High Complexity)
+- Turn 1: Think through and plan the architecture. Identify ALL files needed (e.g. HTML, CSS, JS, tests).
+- Sequential Creation: Write each file completely using write_file in logical dependency order.
+- No Thrashing: Do NOT delete files you just created with `rm`. If adjustments are needed, use edit_file or overwrite directly.
+- Verification: After writing the files, run or verify them using bash (e.g. run test suite, build check, or verify syntax).
+- Early Stopping: Once verified and working, output: DONE: <summary> immediately. Do not run redundant checks.
+"""
+
+GOAL_DRIVEN_RULES = """\
+## Execution & Goal Convergence Rules
+- Efficiency First: Plan your actions to minimize wasted turns. You can execute multiple tool calls in a single turn.
+- No Thrashing: NEVER delete a file (e.g. `rm <file>`) immediately after creating it to start over. Use `edit_file` to modify what needs fixing.
+- Immediate Completion: As soon as your code is written and verified, output `DONE: <summary>`. Do NOT linger or rerun commands that already passed.
+- Targeted Editing: If `edit_file` fails, use `read_file` to inspect the exact lines and whitespace before attempting another edit.
+"""
+
 INTENT_MODIFY = """\
 ## Task Intent: Modify Existing Code
 - Turn 1: Use read_file to read the relevant file(s)
@@ -492,13 +505,19 @@ class PromptBuilder:
             # Inject intent-specific workflow
             intent = getattr(task_context, "intent", "create")
             if intent == "create":
-                prompt += "\n\n" + INTENT_CREATE
+                if complexity in ("medium", "high"):
+                    prompt += "\n\n" + INTENT_CREATE_COMPLEX
+                else:
+                    prompt += "\n\n" + INTENT_CREATE
             elif intent == "modify":
                 prompt += "\n\n" + INTENT_MODIFY
             elif intent == "debug":
                 prompt += "\n\n" + INTENT_DEBUG
             elif intent == "explain":
                 prompt += "\n\n" + INTENT_EXPLAIN
+
+            if complexity in ("medium", "high"):
+                prompt += "\n\n" + GOAL_DRIVEN_RULES
             # qa intent: no extra prompt needed (model answers directly)
 
         memories = format_memories_for_prompt(workdir)

{alpiecode-5.1.1 → alpiecode-6.1.0}/src/codeagent/tools.py

@@ -1,3 +1,5 @@
+_RECENTLY_WRITTEN_FILES = set()
+
 """
 Tool definitions and implementations for AlpieCode.
 
@@ -437,11 +439,11 @@ def _bash(workdir: Path, command: str) -> str:
             shell_cmd,
             cwd=workdir, capture_output=True, text=True,
             stdin=subprocess.DEVNULL,
-            timeout=30, env=env,
+            timeout=120, env=env,
         )
 
-        stdout = _smart_truncate(result.stdout, 6000)
-        stderr = _smart_truncate(result.stderr, 4000)
+        stdout = _smart_truncate(result.stdout, 3000)
+        stderr = _smart_truncate(result.stderr, 2000)
 
         output = json.dumps({
             "stdout": stdout,
@@ -463,6 +465,20 @@ def _bash(workdir: Path, command: str) -> str:
                 )
             output = hint + output
 
+        # ── Anti-thrashing guard: warned if deleting file created in this session ──
+        cmd_parts = command.strip().split()
+        if cmd_parts and cmd_parts[0] in ("rm", "del", "Remove-Item"):
+            for part in cmd_parts[1:]:
+                base_part = Path(part.strip('"\'')).name.lower()
+                if base_part in _RECENTLY_WRITTEN_FILES:
+                    _RECENTLY_WRITTEN_FILES.discard(base_part)
+                    output += (
+                        f"\n\n💡 ADVICE: You deleted '{base_part}' which was created in a recent turn. "
+                        "Avoid deleting and rewriting entire files from scratch. "
+                        "For iterative improvements, use edit_file or overwrite directly with write_file."
+                    )
+                    break
+
         # ── Smart Guardrail 2: Silent success — script ran but no output ──
         elif result.returncode == 0 and not stdout.strip() and not stderr.strip():
             script_file = _extract_script_path(command)
@@ -505,7 +521,8 @@ def _read_file(workdir: Path, path: str, start_line: int = None, end_line: int =
 def _write_file(workdir: Path, path: str, content: str) -> str:
     p = workdir / path
     p.parent.mkdir(parents=True, exist_ok=True)
-    p.write_text(content)
+    p.write_text(content, encoding="utf-8")
+    _RECENTLY_WRITTEN_FILES.add(Path(path).name.lower())
     return f"wrote {len(content)} bytes to {path}"
 
 
@@ -513,13 +530,25 @@ def _edit_file(workdir: Path, path: str, old_str: str, new_str: str) -> str:
     p = workdir / path
     if not p.exists():
         return f"error: file not found: {path}"
-    text = p.read_text()
+    text = p.read_text(encoding="utf-8", errors="replace")
     count = text.count(old_str)
     if count == 0:
-        return "error: old_str not found in file. Make sure it matches exactly (including whitespace)."
+        import difflib
+        lines = text.splitlines()
+        old_lines = [ol.strip() for ol in old_str.splitlines() if ol.strip()]
+        target_line = old_lines[0] if old_lines else old_str.strip()
+        close_matches = difflib.get_close_matches(target_line, [l.strip() for l in lines], n=3, cutoff=0.3)
+        nearby = [(i + 1, l) for i, l in enumerate(lines) if l.strip() in close_matches]
+        hint = ""
+        if nearby:
+            hint = "\nClosest matching lines in file:\n" + "\n".join(f"  Line {n}: {l.strip()[:80]}" for n, l in nearby[:3])
+        return (
+            f"error: old_str not found in file '{path}'. It must match existing text EXACTLY (including whitespace/indentation).{hint}\n"
+            f"→ Action: Run read_file on '{path}' around these lines to see the exact whitespace, then retry edit_file."
+        )
     if count > 1:
-        return f"error: old_str matched {count} times, need exactly 1 match. Use a more specific old_str."
-    p.write_text(text.replace(old_str, new_str, 1))
+        return f"error: old_str matched {count} times in '{path}'. Must match exactly 1 occurrence. Include more surrounding lines in old_str to make it unique."
+    p.write_text(text.replace(old_str, new_str, 1), encoding="utf-8")
     return "edit applied"
 
 
@@ -689,9 +718,10 @@ def _request_user_input(question: str) -> str:
     try:
         from rich.console import Console
         from rich.panel import Panel
+        from rich.text import Text
         console = Console()
         console.print(Panel(
-            f"[bold]{question}[/bold]",
+            Text(question, style="bold"),
             title="❓ Agent needs your input",
             border_style="yellow",
         ))