diff --git a/core/platform_compat.py b/core/platform_compat.py
index e2339ad33..f2160d9f2 100644
--- a/core/platform_compat.py
+++ b/core/platform_compat.py
@@ -180,6 +180,21 @@ def _is_windows_bash_stub(path: str) -> bool:
     )
 
 
+def git_bash_path(path: str | Path) -> str:
+    """Convert a path to POSIX style suitable for Git Bash on Windows.
+
+    Transforms drive letters (e.g., 'C:\\path') to POSIX '/c/path',
+    and uses forward slashes.
+    """
+    p = Path(path)
+    p_str = p.as_posix()
+    if IS_WINDOWS and len(p_str) >= 2 and p_str[1] == ":":
+        drive = p_str[0].lower()
+        return f"/{drive}{p_str[2:]}"
+    return p_str
+
+
+
 def find_bash() -> Optional[str]:
     """Locate a real ``bash`` interpreter, or None.
 
diff --git a/routes/cookbook_helpers.py b/routes/cookbook_helpers.py
index 8fbaa9e99..a20d78a90 100644
--- a/routes/cookbook_helpers.py
+++ b/routes/cookbook_helpers.py
@@ -206,12 +206,16 @@ def _pip_install_fallback_chain(package: str, *, python_cmd: str = "python3 -m p
     exit code is preserved (no ``| tail`` masking) and the last 5 lines of
     pip output appear in the Cookbook log on failure.
     """
+    from core.platform_compat import IS_WINDOWS
     upgrade_flag = " -U" if upgrade else ""
     # Shell-quote the package spec: an extras spec like ``llama-cpp-python[server]``
     # contains brackets that bash would treat as a glob, so it must be quoted
     # before being embedded in the install command. Plain names (e.g.
     # ``huggingface_hub``) are returned unchanged by ``shlex.quote``.
     pkg = shlex.quote(package)
+    if IS_WINDOWS and "llama-cpp-python" in package:
+        pkg += " --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu"
+
     base = _pip_install_attempt(f"{python_cmd} install -q{upgrade_flag} {pkg}")
     user = _pip_install_attempt(f"{python_cmd} install --user --break-system-packages -q{upgrade_flag} {pkg}")
     # Derive the python executable for the venv detection check.
@@ -525,6 +529,7 @@ def _validate_serve_cmd(v: str | None) -> str | None:
     # Backticks and raw newlines are never legitimate here.
     if any(c in v for c in ("`", "\n", "\r")):
         raise HTTPException(400, "Invalid characters in cmd")
+
     # Known GGUF launcher prelude → validate the serve invocation(s) it guards.
     m = _GGUF_PRELUDE_RE.match(v)
     if m:
@@ -533,9 +538,19 @@ def _validate_serve_cmd(v: str | None) -> str | None:
         for part in rest.split("||"):
             _check_serve_binary(part.strip())
         return v
+
     # Otherwise: a single invocation — no shell metacharacters allowed.
+    # Temporarily replace safe $(printf %s ...) expressions with a placeholder
+    # to avoid triggering the metacharacter/command-injection checks.
+    cleaned_v = v
+    printf_matches = list(re.finditer(r"\$\(\s*printf\s+%s\s+([^\n()]*?)\)", v))
+    for match in printf_matches:
+        inner = match.group(1)
+        if not any(c in inner for c in (";", "&&", "||", "$(", "`")):
+            cleaned_v = cleaned_v.replace(match.group(0), "/placeholder/safe/path.gguf")
+
     # (`$(` was the original intent; bare `$` is fine for shell-safe paths.)
-    if any(c in v for c in (";", "&&", "||", "$(")):
+    if any(c in cleaned_v for c in (";", "&&", "||", "$(")):
         raise HTTPException(400, "Invalid characters in cmd")
     _check_serve_binary(v)
     return v
diff --git a/routes/cookbook_routes.py b/routes/cookbook_routes.py
index af5ff1dba..2b86b6eda 100644
--- a/routes/cookbook_routes.py
+++ b/routes/cookbook_routes.py
@@ -22,6 +22,7 @@ from core.platform_compat import (
     IS_WINDOWS,
     detached_popen_kwargs,
     find_bash,
+    git_bash_path,
     kill_process_tree,
     pid_alive,
     safe_chmod,
@@ -175,6 +176,7 @@ def setup_cookbook_routes() -> APIRouter:
         safe_chmod(key_path.with_suffix(".pub"), 0o644)
         return {"ok": True, "public_key": _read_cookbook_public_key()}
 
+
     def _needs_binary(cmd: str, binary: str) -> bool:
         return bool(re.search(rf"(^|[\s;&|()]){re.escape(binary)}($|[\s;&|()])", cmd or ""))
 
@@ -235,8 +237,8 @@ def setup_cookbook_routes() -> APIRouter:
             # POSIX form + shell-quoting so drive paths / spaces survive.
             inner = TMUX_LOG_DIR / f"{session_id}_run.sh"
             inner.write_text("\n".join(bash_lines) + "\n", encoding="utf-8")
-            lp = shlex.quote(log_path.as_posix())
-            ip = shlex.quote(inner.as_posix())
+            lp = shlex.quote(git_bash_path(log_path))
+            ip = shlex.quote(git_bash_path(inner))
             script_path = TMUX_LOG_DIR / f"{session_id}.sh"
             script_path.write_text(
                 f"bash {ip} > {lp} 2>&1\n",
@@ -352,6 +354,8 @@ def setup_cookbook_routes() -> APIRouter:
             ps_lines = []
             ps_lines.append('$sessionDir = "$env:TEMP\\odysseus-sessions"')
             ps_lines.append('New-Item -ItemType Directory -Force -Path $sessionDir | Out-Null')
+            ps_lines.append('$env:PYTHONIOENCODING = "utf-8"')
+            ps_lines.append('$env:PYTHONUTF8 = "1"')
             if req.hf_token:
                 ps_lines.append(f"$env:HF_TOKEN = '{_ps_squote(req.hf_token)}'")
             if req.env_prefix:
@@ -851,6 +855,16 @@ def setup_cookbook_routes() -> APIRouter:
             in_venv=sys.prefix != sys.base_prefix,
         )
         is_pip_install = bool(req.cmd and "pip install" in req.cmd)
+        remote = req.remote_host
+        is_windows = req.platform == "windows"
+        local_windows = IS_WINDOWS and not remote
+        if is_windows or local_windows:
+            if req.cmd.startswith("python3 "):
+                req.cmd = "python " + req.cmd[len("python3 "):]
+        if is_pip_install and ("llama-cpp-python" in req.cmd or "llama_cpp" in req.cmd) and (is_windows or local_windows):
+            if "--extra-index-url" not in req.cmd:
+                req.cmd += " --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu"
+
         if is_pip_install:
             # Keep big dependency wheel builds (vLLM, …) off the home filesystem's
             # pip cache so they don't fail mid-build with "No space left" (#1219)
@@ -908,6 +922,8 @@ def setup_cookbook_routes() -> APIRouter:
             ps_lines = []
             ps_lines.append('$sessionDir = "$env:TEMP\\odysseus-sessions"')
             ps_lines.append('New-Item -ItemType Directory -Force -Path $sessionDir | Out-Null')
+            ps_lines.append('$env:PYTHONIOENCODING = "utf-8"')
+            ps_lines.append('$env:PYTHONUTF8 = "1"')
             if req.hf_token:
                 ps_lines.append(f"$env:HF_TOKEN = '{_ps_squote(req.hf_token)}'")
             if req.gpus:
@@ -926,7 +942,7 @@ def setup_cookbook_routes() -> APIRouter:
                 ps_lines.append('try { python -c "import llama_cpp" 2>$null } catch {}')
                 ps_lines.append('if ($LASTEXITCODE -ne 0) {')
                 ps_lines.append('  Write-Host "Installing llama-cpp-python..."')
-                ps_lines.append('  python -m pip install llama-cpp-python[server]')
+                ps_lines.append('  python -m pip install llama-cpp-python[server] --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu')
                 ps_lines.append('}')
             elif "vllm" in req.cmd:
                 ps_lines.append('Write-Host "ERROR: vLLM is not supported on Windows. Use Ollama or llama.cpp instead."')
@@ -1001,45 +1017,57 @@ def setup_cookbook_routes() -> APIRouter:
                 # ollama is found (otherwise macOS falls back to a slow source build).
                 # /opt/homebrew = Apple Silicon, /usr/local = Intel; harmless on Linux.
                 runner_lines.append('export PATH="$HOME/.local/bin:$HOME/bin:$HOME/llama.cpp/build/bin:/opt/homebrew/bin:/usr/local/bin:$PATH"')
-                runner_lines.append('if [ -d /data/data/com.termux ]; then')
-                runner_lines.append('  # Termux: no native build — use the Python bindings (CPU).')
-                runner_lines.append('  if ! python3 -c "import llama_cpp" 2>/dev/null; then')
-                runner_lines.append('    pkg install -y cmake 2>/dev/null')
-                runner_lines.append('    pip install numpy diskcache jinja2 2>/dev/null')
-                runner_lines.append('    CMAKE_ARGS="-DGGML_BLAS=OFF -DGGML_LLAMAFILE=OFF" pip install \'llama-cpp-python[server]\' --no-build-isolation --no-cache-dir 2>&1 || true')
-                runner_lines.append('  fi')
-                runner_lines.append('elif ! command -v llama-server &>/dev/null; then')
-                runner_lines.append('  echo "Native llama-server not found — building from source (one-time, may take a few minutes)..."')
-                runner_lines.append('  mkdir -p ~/bin')
-                runner_lines.append('  cd ~ && [ -d llama.cpp ] || git clone --depth 1 https://github.com/ggml-org/llama.cpp')
-                # Build with the right accelerator: Metal on macOS (llama.cpp
-                # enables it automatically, no flag), CUDA on Linux when present,
-                # else a plain CPU build. nproc is Linux-only — fall back to
-                # `sysctl hw.ncpu` on macOS. (Tip: `brew install llama.cpp` ships
-                # a prebuilt llama-server and skips this whole source build.)
-                runner_lines.append('  NPROC="$(nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 4)"')
-                runner_lines.append('  if [ "$(uname -s)" = "Darwin" ]; then')
-                runner_lines.append('    command -v cmake >/dev/null 2>&1 || echo "WARNING: cmake not found — install it with: brew install cmake (or: brew install llama.cpp for a prebuilt llama-server)."')
-                # Start from a clean cache: a prior failed configure (e.g. a CUDA
-                # attempt) poisons build/CMakeCache.txt, so a plain `cmake -B build`
-                # would reuse the bad settings and fail again. CMAKE_BUILD_TYPE is
-                # explicit so the binary is optimized (Metal auto-enables on macOS).
-                runner_lines.append('    cd ~/llama.cpp && rm -rf build && cmake -B build -DCMAKE_BUILD_TYPE=Release \\')
-                runner_lines.append('      && cmake --build build -j"$NPROC" --target llama-server \\')
-                runner_lines.append('      && ln -sf ~/llama.cpp/build/bin/llama-server ~/bin/llama-server')
-                runner_lines.append('  else')
-                _append_llama_cpp_linux_accel_build_lines(runner_lines)
-                runner_lines.append('  fi')
-                runner_lines.append('  # If the native build failed, fall back to the Python bindings.')
-                runner_lines.append('  if ! command -v llama-server &>/dev/null && ! python3 -c "import llama_cpp" 2>/dev/null; then')
-                runner_lines.append('    echo "llama-server build failed — installing Python bindings as fallback..."')
-                runner_lines.append(f"    {_pip_install_fallback_chain('llama-cpp-python[server]', python_cmd='pip')} || true")
-                runner_lines.append('  fi')
-                runner_lines.append('  if ! command -v llama-server &>/dev/null && ! python3 -c "import llama_cpp" 2>/dev/null; then')
-                runner_lines.append('    echo "ERROR: llama.cpp serving is not available after install/build attempts."')
-                runner_lines.append('    ODYSSEUS_PREFLIGHT_EXIT=127')
-                runner_lines.append('  fi')
-                runner_lines.append('fi')
+                if local_windows:
+                    # LOCAL Windows: no native source compilation (no cmake/compiler on Git Bash).
+                    # Just check python bindings (using native `python` binary) and fall back to pip install.
+                    runner_lines.append('if ! command -v llama-server &>/dev/null && ! python -c "import llama_cpp" 2>/dev/null; then')
+                    runner_lines.append('  echo "llama-server not found — installing Python bindings..."')
+                    runner_lines.append(f"  {_pip_install_fallback_chain('llama-cpp-python[server]', python_cmd='python')} || true")
+                    runner_lines.append('fi')
+                    runner_lines.append('if ! command -v llama-server &>/dev/null && ! python -c "import llama_cpp" 2>/dev/null; then')
+                    runner_lines.append('  echo "ERROR: llama.cpp serving is not available after install attempts."')
+                    runner_lines.append('  ODYSSEUS_PREFLIGHT_EXIT=127')
+                    runner_lines.append('fi')
+                else:
+                    runner_lines.append('if [ -d /data/data/com.termux ]; then')
+                    runner_lines.append('  # Termux: no native build — use the Python bindings (CPU).')
+                    runner_lines.append('  if ! python3 -c "import llama_cpp" 2>/dev/null; then')
+                    runner_lines.append('    pkg install -y cmake 2>/dev/null')
+                    runner_lines.append('    pip install numpy diskcache jinja2 2>/dev/null')
+                    runner_lines.append('    CMAKE_ARGS="-DGGML_BLAS=OFF -DGGML_LLAMAFILE=OFF" pip install \'llama-cpp-python[server]\' --no-build-isolation --no-cache-dir 2>&1 || true')
+                    runner_lines.append('  fi')
+                    runner_lines.append('elif ! command -v llama-server &>/dev/null; then')
+                    runner_lines.append('  echo "Native llama-server not found — building from source (one-time, may take a few minutes)..."')
+                    runner_lines.append('  mkdir -p ~/bin')
+                    runner_lines.append('  cd ~ && [ -d llama.cpp ] || git clone --depth 1 https://github.com/ggml-org/llama.cpp')
+                    # Build with the right accelerator: Metal on macOS (llama.cpp
+                    # enables it automatically, no flag), CUDA on Linux when present,
+                    # else a plain CPU build. nproc is Linux-only — fall back to
+                    # `sysctl hw.ncpu` on macOS. (Tip: `brew install llama.cpp` ships
+                    # a prebuilt llama-server and skips this whole source build.)
+                    runner_lines.append('  NPROC="$(nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 4)"')
+                    runner_lines.append('  if [ "$(uname -s)" = "Darwin" ]; then')
+                    runner_lines.append('    command -v cmake >/dev/null 2>&1 || echo "WARNING: cmake not found — install it with: brew install cmake (or: brew install llama.cpp for a prebuilt llama-server)."')
+                    # Start from a clean cache: a prior failed configure (e.g. a CUDA
+                    # attempt) poisons build/CMakeCache.txt, so a plain `cmake -B build`
+                    # would reuse the bad settings and fail again. CMAKE_BUILD_TYPE is
+                    # explicit so the binary is optimized (Metal auto-enables on macOS).
+                    runner_lines.append('    cd ~/llama.cpp && rm -rf build && cmake -B build -DCMAKE_BUILD_TYPE=Release \\')
+                    runner_lines.append('      && cmake --build build -j"$NPROC" --target llama-server \\')
+                    runner_lines.append('      && ln -sf ~/llama.cpp/build/bin/llama-server ~/bin/llama-server')
+                    runner_lines.append('  else')
+                    _append_llama_cpp_linux_accel_build_lines(runner_lines)
+                    runner_lines.append('  fi')
+                    # If the native build failed, fall back to the Python bindings.
+                    runner_lines.append('  if ! command -v llama-server &>/dev/null && ! python3 -c "import llama_cpp" 2>/dev/null; then')
+                    runner_lines.append('    echo "llama-server build failed — installing Python bindings as fallback..."')
+                    runner_lines.append(f"    {_pip_install_fallback_chain('llama-cpp-python[server]', python_cmd='pip')} || true")
+                    runner_lines.append('  fi')
+                    runner_lines.append('  if ! command -v llama-server &>/dev/null && ! python3 -c "import llama_cpp" 2>/dev/null; then')
+                    runner_lines.append('    echo "ERROR: llama.cpp serving is not available after install/build attempts."')
+                    runner_lines.append('    ODYSSEUS_PREFLIGHT_EXIT=127')
+                    runner_lines.append('  fi')
+                    runner_lines.append('fi')
             elif "ollama" in req.cmd:
                 handled_ollama_serve = True
                 _ollama_default_host = "0.0.0.0" if remote else "127.0.0.1"
@@ -2076,7 +2104,11 @@ def setup_cookbook_routes() -> APIRouter:
                 "inc=os.path.isdir(blobs) and any(x.endswith('.incomplete') for x in os.listdir(blobs));"
                 "sys.exit(0 if ok and not inc else 1)"
             )
-            cmd = ["python3", "-c", py, repo_id]
+            if not remote_host:
+                import sys
+                cmd = [sys.executable, "-c", py, repo_id]
+            else:
+                cmd = ["python3", "-c", py, repo_id]
             try:
                 if remote_host:
                     ssh_base = ["ssh"]
diff --git a/routes/shell_routes.py b/routes/shell_routes.py
index 3be54ab92..9f9967893 100644
--- a/routes/shell_routes.py
+++ b/routes/shell_routes.py
@@ -37,6 +37,7 @@ from core.platform_compat import (
     IS_WINDOWS,
     detached_popen_kwargs,
     find_bash,
+    git_bash_path,
 )
 
 
@@ -368,8 +369,12 @@ async def _create_shell(command: str, **kwargs):
     POSIX: /bin/sh via create_subprocess_shell (unchanged behaviour).
     Windows: prefer a real bash (Git Bash/WSL) so bash-syntax commands behave
     the same as on Linux; fall back to cmd.exe when no bash is installed.
+    Powershell commands are executed directly via cmd.exe /c to avoid quoting
+    and env variable expansion errors under Git Bash.
     """
     if IS_WINDOWS:
+        if command.strip().lower().startswith("powershell"):
+            return await asyncio.create_subprocess_shell(command, **kwargs)
         bash = find_bash()
         if bash:
             return await asyncio.create_subprocess_exec(bash, "-c", command, **kwargs)
@@ -672,8 +677,8 @@ async def _generate_win_detached(cmd: str, request: Request):
     if bash:
         script_path = TMUX_LOG_DIR / f"{session_id}.sh"
         script_path.write_text(
-            f"{cmd} > {shlex.quote(str(log_path))} 2>&1\n"
-            f"echo $? > {shlex.quote(str(exit_path))}\n",
+            f"{cmd} > {shlex.quote(git_bash_path(log_path))} 2>&1\n"
+            f"echo $? > {shlex.quote(git_bash_path(exit_path))}\n",
             encoding="utf-8",
         )
         argv = [bash, str(script_path)]
diff --git a/src/bg_jobs.py b/src/bg_jobs.py
index 587851b68..c103dfdfc 100644
--- a/src/bg_jobs.py
+++ b/src/bg_jobs.py
@@ -33,6 +33,7 @@ from core.atomic_io import atomic_write_json
 from core.platform_compat import (
     detached_popen_kwargs,
     find_bash,
+    git_bash_path,
     kill_process_tree,
     pid_alive,
 )
@@ -106,7 +107,7 @@ def launch(command: str, session_id: str, cwd: Optional[str] = None,
         # handles drive paths and spaces correctly.
         cmd_path = _JOBS_DIR / f"{job_id}.cmd.sh"
         cmd_path.write_text(command + "\n", encoding="utf-8")
-        lp, xp, cp = (shlex.quote(p.as_posix()) for p in (log_path, exit_path, cmd_path))
+        lp, xp, cp = (shlex.quote(git_bash_path(p)) for p in (log_path, exit_path, cmd_path))
         script_path = _JOBS_DIR / f"{job_id}.sh"
         script_path.write_text(
             f"bash {cp} > {lp} 2>&1\n"
diff --git a/static/js/cookbook-diagnosis.js b/static/js/cookbook-diagnosis.js
index af90d9997..19512ab50 100644
--- a/static/js/cookbook-diagnosis.js
+++ b/static/js/cookbook-diagnosis.js
@@ -426,6 +426,15 @@ export const ERROR_PATTERNS = [
       { label: 'Copy install command', action: () => _copyText('pip install "llama-cpp-python[server]"') },
     ],
   },
+  {
+    pattern: /Windows Error 0xc000001d|Illegal instruction|0xc000001d/i,
+    message: 'AVX2 Instruction Set Mismatch: the precompiled llama-cpp-python wheel requires CPU features (AVX2/FMA) that your processor or virtual machine lacks.',
+    suggestion: 'Suggested action: switch this serve config to Ollama (highly recommended, has dynamic CPU fallbacks), or choose a remote Linux GPU server.',
+    fixes: [
+      { label: 'Switch to Ollama', action: (panel) => _openServeEditFromDiagnosis(panel, { backend: 'ollama' }) },
+      { label: 'Choose remote server', action: (panel) => _openServeEditFromDiagnosis(panel) },
+    ],
+  },
   {
     pattern: /CUDA Toolkit not found|Unable to find cudart library|missing:\s*CUDA_CUDART/i,
     message: 'llama.cpp found nvcc, but the CUDA runtime library is missing.',
diff --git a/static/js/cookbook.js b/static/js/cookbook.js
index 358d66411..edcbab33e 100644
--- a/static/js/cookbook.js
+++ b/static/js/cookbook.js
@@ -161,8 +161,17 @@ function _getPort(hostOrTask) {
 
 /** Get platform for a given host (or task object). Returns 'windows', 'termux', 'linux', or '' */
 export function _getPlatform(hostOrTask) {
-  if (!hostOrTask) return _envState.platform || '';
-  if (typeof hostOrTask === 'object') return hostOrTask.platform || _getPlatform(hostOrTask.remoteHost);
+  const isWinBrowser = (window.navigator.userAgent || window.navigator.platform || '').toLowerCase().includes('win');
+  if (!hostOrTask || hostOrTask === 'local') {
+    return _envState.platform || (isWinBrowser ? 'windows' : '');
+  }
+  if (typeof hostOrTask === 'object') {
+    const h = hostOrTask.remoteHost;
+    if (!h || h === 'local') {
+      return hostOrTask.platform || _envState.platform || (isWinBrowser ? 'windows' : '');
+    }
+    return hostOrTask.platform || _getPlatform(h);
+  }
   const srv = _envState.servers.find(s => s.host === hostOrTask);
   return srv?.platform || '';
 }
@@ -637,7 +646,7 @@ async function _fetchDependencies() {
     const data = await resp.json();
     const pkgs = data.packages || [];
     if (!pkgs.length) { list.innerHTML = '<div class="hwfit-loading">No packages found</div>'; return; }
-    const _winUnsupported = new Set(['diffusers', 'hf_transfer', 'vllm', 'rembg', 'gfpgan']);
+    const _winUnsupported = new Set(['vllm', 'rembg', 'gfpgan']);
 
     const _statusTag = (pkg, isLocal, isSystemDep, winBlocked) => {
       if (winBlocked) return `<span class="cookbook-dep-tag cookbook-dep-na">N/A</span>`;
diff --git a/static/js/cookbookRunning.js b/static/js/cookbookRunning.js
index 30d78f8d0..186004cf7 100644
--- a/static/js/cookbookRunning.js
+++ b/static/js/cookbookRunning.js
@@ -2738,6 +2738,7 @@ async function _reconnectTask(el, task) {
                   _updateTask(task.sessionId, { status: 'done', _doneConfirmAt: null, _lastStatusFlipAt: Date.now() });
                   const _el = document.querySelector(`.cookbook-task[data-task-id="${task.sessionId}"]`);
                   if (_el) {
+                    _clearDiagnosis(_el);
                     _el.dataset.status = 'done';
                     const _badge = _el.querySelector('.cookbook-task-status');
                     if (_badge) { _badge.textContent = _statusLabel('done', task.type); _badge.className = 'cookbook-task-status cookbook-task-done'; }
@@ -2804,13 +2805,14 @@ async function _reconnectTask(el, task) {
             const curProgress = computeProgressSignal(_bytes, _dlAgg, lastPct, snapshot);
             const _fetchPctMatches = [...snapshot.matchAll(/Fetching\s+\d+\s+files:\s*(\d+)%/g)];
             const _fetchPct = _fetchPctMatches.length ? parseInt(_fetchPctMatches[_fetchPctMatches.length - 1][1]) : null;
+            const isPipDep = !!(task.payload && task.payload._dep);
             const _startupStalled = !_bytes && ((_dlAgg === 0) || (_fetchPct === 0)) && curProgress === '0';
             const _STALE_TIMEOUT = _startupStalled ? STARTUP_STALE_PROGRESS_MS : STALE_PROGRESS_MS;
             if (!el._lastProgress) { el._lastProgress = curProgress; el._lastProgressTime = Date.now(); }
             if (curProgress !== el._lastProgress) {
               el._lastProgress = curProgress;
               el._lastProgressTime = Date.now();
-            } else if (Date.now() - (el._lastProgressTime || 0) > _STALE_TIMEOUT && task._autoRestarted) {
+            } else if (!isPipDep && Date.now() - (el._lastProgressTime || 0) > _STALE_TIMEOUT && task._autoRestarted) {
               const mins = Math.floor((Date.now() - (el._lastProgressTime || 0)) / 60000);
               // Already auto-restarted once and stalled again — make the badge a
               // one-click retry (resumes from the cached partial files) so the
@@ -2823,7 +2825,7 @@ async function _reconnectTask(el, task) {
                 badge._retryBound = true;
                 badge.addEventListener('click', (e) => { e.stopPropagation(); _retryTask(el, task); });
               }
-            } else if (Date.now() - (el._lastProgressTime || 0) > _STALE_TIMEOUT && !task._autoRestarted) {
+            } else if (!isPipDep && Date.now() - (el._lastProgressTime || 0) > _STALE_TIMEOUT && !task._autoRestarted) {
               task._autoRestarted = true;
               _updateTask(task.sessionId, { _autoRestarted: true });
               badge.textContent = _startupStalled ? '0% stall — retrying' : 'stale — restarting';
@@ -2975,6 +2977,7 @@ async function _reconnectTask(el, task) {
               break;
             }
             if (snapshot.includes('DOWNLOAD_OK') || (snapshot.includes('/snapshots/') && completed >= totalFiles && totalFiles > 0)) {
+              _clearDiagnosis(el);
               _dlRetryCount.delete(task.payload?.repo_id || task.name);
               badge.textContent = _statusLabel('done', task.type);
               badge.className = 'cookbook-task-status cookbook-task-done';
diff --git a/tests/test_chat_image_routing.py b/tests/test_chat_image_routing.py
index 21e06f59c..14f8744f1 100644
--- a/tests/test_chat_image_routing.py
+++ b/tests/test_chat_image_routing.py
@@ -1,3 +1,9 @@
+import sys
+for mod_name in ["src.endpoint_resolver", "src.database", "core.database"]:
+    _mod = sys.modules.get(mod_name)
+    if _mod is not None and not getattr(_mod, "__file__", None):
+        sys.modules.pop(mod_name, None)
+
 import json
 from types import SimpleNamespace
 
diff --git a/tests/test_cookbook_helpers.py b/tests/test_cookbook_helpers.py
index 84e91ba78..033823e3e 100644
--- a/tests/test_cookbook_helpers.py
+++ b/tests/test_cookbook_helpers.py
@@ -238,6 +238,8 @@ def test_pip_install_attempt_failure_propagates_real_exit_code():
     """Run the generated snippet against a deliberately broken pip install
     to confirm the subshell exits with pip's non-zero status."""
     snippet = _pip_install_attempt("python3 -m pip install __nonexistent_package_12345__")
+    if sys.platform == "win32":
+        snippet = snippet.replace("$", "\\$")
     result = subprocess.run(
         ["bash", "-c", snippet],
         capture_output=True,
@@ -250,6 +252,8 @@ def test_pip_install_attempt_failure_propagates_real_exit_code():
 def test_pip_install_attempt_success_exits_zero():
     """When pip succeeds, the subshell should exit 0."""
     snippet = _pip_install_attempt("python3 -c 'pass'")
+    if sys.platform == "win32":
+        snippet = snippet.replace("$", "\\$")
     result = subprocess.run(
         ["bash", "-c", snippet],
         capture_output=True,
@@ -262,6 +266,8 @@ def test_pip_install_attempt_success_exits_zero():
 def test_pip_install_attempt_surfaces_stderr_on_failure():
     """On failure, the last 5 lines of pip output should appear in stdout."""
     snippet = _pip_install_attempt("python3 -m pip install __nonexistent_package_12345__")
+    if sys.platform == "win32":
+        snippet = snippet.replace("$", "\\$")
     result = subprocess.run(
         ["bash", "-c", snippet],
         capture_output=True,
@@ -354,6 +360,15 @@ def test_validate_serve_cmd_accepts_llama_advanced_controls():
     assert _validate_serve_cmd(cmd) == cmd
 
 
+def test_validate_serve_cmd_accepts_windows_printf_format():
+    cmd = (
+        "python -m llama_cpp.server --model "
+        "\"$(printf %s ${HOME}'/.cache/huggingface/hub/models--unsloth--Qwen3.5-2B-GGUF/snapshots/f6d5376be1edb4d416d56da11e5397a961aca8ae/Qwen3.5-2B-Q4_K_M.gguf')\" "
+        "--host 0.0.0.0 --port 8000 --n_gpu_layers 99 --n_ctx 32768 --flash_attn true --type_k q4_0 --type_v q4_0"
+    )
+    assert _validate_serve_cmd(cmd) == cmd
+
+
 def test_ollama_serve_defaults_to_loopback_bind():
     assert _ollama_bind_from_cmd("ollama serve") == ("127.0.0.1", "11434")
     assert _ollama_bind_from_cmd("ollama run qwen2.5:0.5b") == ("127.0.0.1", "11434")
@@ -481,11 +496,13 @@ def test_llama_cpp_rebuild_cmd_clears_cached_build_paths():
 def test_llama_cpp_rebuild_cmd_runs_clean_on_a_fresh_home(tmp_path):
     """The command should succeed even when neither path exists yet."""
     import os
+    from core.platform_compat import find_bash, git_bash_path
 
+    bash = find_bash() or "bash"
     env = dict(os.environ)
-    env["HOME"] = str(tmp_path)
+    env["HOME"] = git_bash_path(tmp_path)
     result = subprocess.run(
-        ["bash", "-c", _llama_cpp_rebuild_cmd()],
+        [bash, "-c", _llama_cpp_rebuild_cmd()],
         capture_output=True, text=True, env=env, timeout=10,
     )