Merge origin/cache

ai_client.py

@@ -17,7 +17,9 @@ import time
 import datetime
 import hashlib
 import difflib
+import threading
 from pathlib import Path
+import os
 import file_cache
 import mcp_client
 import anthropic
@@ -53,6 +55,8 @@ _GEMINI_CACHE_TTL = 3600
 
 _anthropic_client = None
 _anthropic_history: list[dict] = []
+_anthropic_history_lock = threading.Lock()
+_send_lock = threading.Lock()
 
 # Injected by gui.py - called when AI wants to run a command.
 # Signature: (script: str, base_dir: str) -> str | None
@@ -69,6 +73,10 @@ tool_log_callback = None
 # Increased to allow thorough code exploration before forcing a summary
 MAX_TOOL_ROUNDS = 10
 
+# Maximum cumulative bytes of tool output allowed per send() call.
+# Prevents unbounded memory growth during long tool-calling loops.
+_MAX_TOOL_OUTPUT_BYTES = 500_000
+
 # Maximum characters per text chunk sent to Anthropic.
 # Kept well under the ~200k token API limit.
 _ANTHROPIC_CHUNK_SIZE = 120_000
@@ -130,8 +138,18 @@ def clear_comms_log():
 
 
 def _load_credentials() -> dict:
-    with open("credentials.toml", "rb") as f:
+    cred_path = os.environ.get("SLOP_CREDENTIALS", "credentials.toml")
-        return tomllib.load(f)
+    try:
+        with open(cred_path, "rb") as f:
+            return tomllib.load(f)
+    except FileNotFoundError:
+        raise FileNotFoundError(
+            f"Credentials file not found: {cred_path}\n"
+            f"Create a credentials.toml with:\n"
+            f"  [gemini]\n  api_key = \"your-key\"\n"
+            f"  [anthropic]\n  api_key = \"your-key\"\n"
+            f"Or set SLOP_CREDENTIALS env var to a custom path."
+        )
 
 
 # ------------------------------------------------------------------ provider errors
@@ -246,7 +264,8 @@ def reset_session():
     _gemini_cache_md_hash = None
     _gemini_cache_created_at = None
     _anthropic_client = None
-    _anthropic_history = []
+    with _anthropic_history_lock:
+        _anthropic_history = []
     _CACHED_ANTHROPIC_TOOLS = None
     file_cache.reset_client()
 
@@ -652,6 +671,7 @@ def _send_gemini(md_content: str, user_message: str, base_dir: str,
 
         _append_comms("OUT", "request", {"message": f"[ctx {len(md_content)} + msg {len(user_message)}]"})
         payload, all_text = user_message, []
+        _cumulative_tool_bytes = 0
 
         # Strip stale file refreshes and truncate old tool outputs ONCE before
         # entering the tool loop (not per-round — history entries don't change).
@@ -701,11 +721,11 @@ def _send_gemini(md_content: str, user_message: str, base_dir: str,
                         if not hist: break
                         for p in hist[0].parts:
                             if hasattr(p, "text") and p.text:
-                                saved += len(p.text) // 4
+                                saved += int(len(p.text) / _CHARS_PER_TOKEN)
                             elif hasattr(p, "function_response") and p.function_response:
                                 r = getattr(p.function_response, "response", {})
                                 if isinstance(r, dict):
-                                    saved += len(str(r.get("output", ""))) // 4
+                                    saved += int(len(str(r.get("output", ""))) / _CHARS_PER_TOKEN)
                         hist.pop(0)
                         dropped += 1
                     total_in -= max(saved, 200)
@@ -736,10 +756,17 @@ def _send_gemini(md_content: str, user_message: str, base_dir: str,
                     if r_idx == MAX_TOOL_ROUNDS: out += "\n\n[SYSTEM: MAX ROUNDS. PROVIDE FINAL ANSWER.]"
 
                 out = _truncate_tool_output(out)
+                _cumulative_tool_bytes += len(out)
                 f_resps.append(types.Part.from_function_response(name=name, response={"output": out}))
                 log.append({"tool_use_id": name, "content": out})
                 events.emit("tool_execution", payload={"status": "completed", "tool": name, "result": out, "round": r_idx})
-            
+
+            if _cumulative_tool_bytes > _MAX_TOOL_OUTPUT_BYTES:
+                f_resps.append(types.Part.from_text(
+                    f"SYSTEM WARNING: Cumulative tool output exceeded {_MAX_TOOL_OUTPUT_BYTES // 1000}KB budget. Provide your final answer now."
+                ))
+                _append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
+
             _append_comms("OUT", "tool_result_send", {"results": log})
             payload = f_resps
             
@@ -1046,6 +1073,7 @@ def _send_anthropic(md_content: str, user_message: str, base_dir: str, file_item
         })
 
         all_text_parts = []
+        _cumulative_tool_bytes = 0
 
         # We allow MAX_TOOL_ROUNDS, plus 1 final loop to get the text synthesis
         for round_idx in range(MAX_TOOL_ROUNDS + 2):
@@ -1132,10 +1160,12 @@ def _send_anthropic(md_content: str, user_message: str, base_dir: str, file_item
                     _append_comms("OUT", "tool_call", {"name": b_name, "id": b_id, "args": b_input})
                     output = mcp_client.dispatch(b_name, b_input)
                     _append_comms("IN", "tool_result", {"name": b_name, "id": b_id, "output": output})
+                    truncated = _truncate_tool_output(output)
+                    _cumulative_tool_bytes += len(truncated)
                     tool_results.append({
                         "type":        "tool_result",
                         "tool_use_id": b_id,
-                        "content":     _truncate_tool_output(output),
+                        "content":     truncated,
                     })
                     events.emit("tool_execution", payload={"status": "completed", "tool": b_name, "result": output, "round": round_idx})
                 elif b_name == TOOL_NAME:
@@ -1151,13 +1181,22 @@ def _send_anthropic(md_content: str, user_message: str, base_dir: str, file_item
                         "id":     b_id,
                         "output": output,
                     })
+                    truncated = _truncate_tool_output(output)
+                    _cumulative_tool_bytes += len(truncated)
                     tool_results.append({
                         "type":        "tool_result",
                         "tool_use_id": b_id,
-                        "content":     _truncate_tool_output(output),
+                        "content":     truncated,
                     })
                     events.emit("tool_execution", payload={"status": "completed", "tool": b_name, "result": output, "round": round_idx})
 
+            if _cumulative_tool_bytes > _MAX_TOOL_OUTPUT_BYTES:
+                tool_results.append({
+                    "type": "text",
+                    "text": f"SYSTEM WARNING: Cumulative tool output exceeded {_MAX_TOOL_OUTPUT_BYTES // 1000}KB budget. Provide your final answer now."
+                })
+                _append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
+
             # Refresh file context after tool calls — only inject CHANGED files
             if file_items:
                 file_items, changed = _reread_file_items(file_items)
@@ -1220,11 +1259,12 @@ def send(
     discussion_history : discussion history text (used by Gemini to inject as
                          conversation message instead of caching it)
     """
-    if _provider == "gemini":
+    with _send_lock:
-        return _send_gemini(md_content, user_message, base_dir, file_items, discussion_history)
+        if _provider == "gemini":
-    elif _provider == "anthropic":
+            return _send_gemini(md_content, user_message, base_dir, file_items, discussion_history)
-        return _send_anthropic(md_content, user_message, base_dir, file_items, discussion_history)
+        elif _provider == "anthropic":
-    raise ValueError(f"unknown provider: {_provider}")
+            return _send_anthropic(md_content, user_message, base_dir, file_items, discussion_history)
+        raise ValueError(f"unknown provider: {_provider}")
 
 def get_history_bleed_stats() -> dict:
     """
@@ -1232,7 +1272,9 @@ def get_history_bleed_stats() -> dict:
     """
     if _provider == "anthropic":
         # For Anthropic, we have a robust estimator
-        current_tokens = _estimate_prompt_tokens([], _anthropic_history)
+        with _anthropic_history_lock:
+            history_snapshot = list(_anthropic_history)
+        current_tokens = _estimate_prompt_tokens([], history_snapshot)
         limit_tokens = _ANTHROPIC_MAX_PROMPT_TOKENS
         percentage = (current_tokens / limit_tokens) * 100 if limit_tokens > 0 else 0
         return {

gui_2.py

@@ -153,6 +153,7 @@ class App:
         self.last_file_items: list = []
         
         self.send_thread: threading.Thread | None = None
+        self._send_thread_lock = threading.Lock()
         self.models_thread: threading.Thread | None = None
 
         _default_windows = {
@@ -232,6 +233,10 @@ class App:
         self.perf_history = {"frame_time": [0.0]*100, "fps": [0.0]*100, "cpu": [0.0]*100, "input_lag": [0.0]*100}
         self._perf_last_update = 0.0
 
+        # Auto-save timer (every 60s)
+        self._autosave_interval = 60.0
+        self._last_autosave = time.time()
+
         session_logger.open_session()
         ai_client.set_provider(self.current_provider, self.current_model)
         ai_client.confirm_and_run_callback = self._confirm_and_run
@@ -625,6 +630,18 @@ class App:
         # Process GUI task queue
         self._process_pending_gui_tasks()
 
+        # Auto-save (every 60s)
+        now = time.time()
+        if now - self._last_autosave >= self._autosave_interval:
+            self._last_autosave = now
+            try:
+                self._flush_to_project()
+                self._save_active_project()
+                self._flush_to_config()
+                save_config(self.config)
+            except Exception:
+                pass  # silent — don't disrupt the GUI loop
+
         # Sync pending comms
         with self._pending_comms_lock:
             for c in self._pending_comms:
@@ -1109,9 +1126,21 @@ class App:
                     imgui.separator()
 
                 ch, self.ui_ai_input = imgui.input_text_multiline("##ai_in", self.ui_ai_input, imgui.ImVec2(-1, -40))
+
+                # Keyboard shortcuts
+                io = imgui.get_io()
+                ctrl_enter = io.key_ctrl and imgui.is_key_pressed(imgui.Key.enter)
+                ctrl_l = io.key_ctrl and imgui.is_key_pressed(imgui.Key.l)
+                if ctrl_l:
+                    self.ui_ai_input = ""
+
                 imgui.separator()
-                if imgui.button("Gen + Send"):
+                send_busy = False
-                    if not (self.send_thread and self.send_thread.is_alive()):
+                with self._send_thread_lock:
+                    if self.send_thread and self.send_thread.is_alive():
+                        send_busy = True
+                if imgui.button("Gen + Send") or ctrl_enter:
+                    if not send_busy:
                         try:
                             md, path, file_items, stable_md, disc_text = self._do_generate()
                             self.last_md = md
@@ -1127,10 +1156,7 @@ class App:
                             ai_client.set_custom_system_prompt("\n\n".join(csp))
                             ai_client.set_model_params(self.temperature, self.max_tokens, self.history_trunc_limit)
                             ai_client.set_agent_tools(self.ui_agent_tools)
-                            # For Gemini: send stable_md (no history) as cached context,
+                            send_md = stable_md
-                            # and disc_text separately as conversation history.
-                            # For Anthropic: send full md (with history) as before.
-                            send_md = stable_md  # No history in cached context for either provider
                             send_disc = disc_text
 
                             def do_send():
@@ -1159,9 +1185,10 @@ class App:
                                     if self.ui_auto_add_history:
                                         with self._pending_history_adds_lock:
                                             self._pending_history_adds.append({"role": "System", "content": self.ai_response, "collapsed": False, "ts": project_manager.now_ts()})
-                                            
+
-                            self.send_thread = threading.Thread(target=do_send, daemon=True)
+                            with self._send_thread_lock:
-                            self.send_thread.start()
+                                self.send_thread = threading.Thread(target=do_send, daemon=True)
+                                self.send_thread.start()
                 imgui.same_line()
                 if imgui.button("MD Only"):
                     try:

mcp_client.py

@@ -65,7 +65,10 @@ def configure(file_items: list[dict], extra_base_dirs: list[str] | None = None):
     for item in file_items:
         p = item.get("path")
         if p is not None:
-            rp = Path(p).resolve()
+            try:
+                rp = Path(p).resolve(strict=True)
+            except (OSError, ValueError):
+                rp = Path(p).resolve()
             _allowed_paths.add(rp)
             _base_dirs.add(rp.parent)
 
@@ -82,8 +85,13 @@ def _is_allowed(path: Path) -> bool:
     A path is allowed if:
       - it is explicitly in _allowed_paths, OR
       - it is contained within (or equal to) one of the _base_dirs
+    All paths are resolved (follows symlinks) before comparison to prevent
+    symlink-based path traversal.
     """
-    rp = path.resolve()
+    try:
+        rp = path.resolve(strict=True)
+    except (OSError, ValueError):
+        rp = path.resolve()
     if rp in _allowed_paths:
         return True
     for bd in _base_dirs:
@@ -104,7 +112,10 @@ def _resolve_and_check(raw_path: str) -> tuple[Path | None, str]:
         p = Path(raw_path)
         if not p.is_absolute() and _primary_base_dir:
             p = _primary_base_dir / p
-        p = p.resolve()
+        try:
+            p = p.resolve(strict=True)
+        except (OSError, ValueError):
+            p = p.resolve()
     except Exception as e:
         return None, f"ERROR: invalid path '{raw_path}': {e}"
     if not _is_allowed(p):
@@ -269,7 +280,8 @@ def web_search(query: str) -> str:
     url = "https://html.duckduckgo.com/html/?q=" + urllib.parse.quote(query)
     req = urllib.request.Request(url, headers={'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64)'})
     try:
-        html = urllib.request.urlopen(req, timeout=10).read().decode('utf-8', errors='ignore')
+        with urllib.request.urlopen(req, timeout=10) as resp:
+            html = resp.read().decode('utf-8', errors='ignore')
         parser = _DDGParser()
         parser.feed(html)
         if not parser.results:
@@ -292,7 +304,8 @@ def fetch_url(url: str) -> str:
 
     req = urllib.request.Request(url, headers={'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64)'})
     try:
-        html = urllib.request.urlopen(req, timeout=10).read().decode('utf-8', errors='ignore')
+        with urllib.request.urlopen(req, timeout=10) as resp:
+            html = resp.read().decode('utf-8', errors='ignore')
         parser = _TextExtractor()
         parser.feed(html)
         full_text = " ".join(parser.text)

session_logger.py

@@ -26,6 +26,7 @@ scripts/generated/
 Where <ts> = YYYYMMDD_HHMMSS of when this session was started.
 """
 
+import atexit
 import datetime
 import json
 import threading
@@ -71,6 +72,8 @@ def open_session():
     _tool_fh.write(f"# Tool-call log — session {_ts}\n\n")
     _tool_fh.flush()
 
+    atexit.register(close_session)
+
 
 def close_session():
     """Flush and close both log files.  Called on clean exit (optional)."""