tau-coding-agent 0.7.7__tar.gz → 0.7.8__tar.gz

{tau_coding_agent-0.7.7/tau_coding_agent.egg-info → tau_coding_agent-0.7.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tau-coding-agent
-Version: 0.7.7
+Version: 0.7.8
 Summary: A self-extensible agent CLI with terminal UI, multi-provider LLM support, session management with branching, and a plugin system for tools and commands
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
@@ -141,11 +141,12 @@ Full flag list, environment variables, and exit codes: [CLI reference](docs/cli-
 ### Subcommands
 
 ```bash
-tau auth        # Manage provider credentials (login/logout, list)
-tau install      # Install a package (extension/skill/theme)
-tau remove        # Remove an installed package
-tau list           # List installed packages
-tau update          # Update installed packages
+tau auth      # Manage provider credentials (login/logout, list)
+tau doctor    # Diagnose config, auth, models, extensions, sessions, packages (--fix to repair)
+tau install   # Install a package (extension/skill/theme)
+tau remove    # Remove an installed package
+tau list      # List installed packages
+tau update    # Update installed packages
 ```
 
 ### Interactive slash commands

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/README.md

@@ -102,11 +102,12 @@ Full flag list, environment variables, and exit codes: [CLI reference](docs/cli-
 ### Subcommands
 
 ```bash
-tau auth        # Manage provider credentials (login/logout, list)
-tau install      # Install a package (extension/skill/theme)
-tau remove        # Remove an installed package
-tau list           # List installed packages
-tau update          # Update installed packages
+tau auth      # Manage provider credentials (login/logout, list)
+tau doctor    # Diagnose config, auth, models, extensions, sessions, packages (--fix to repair)
+tau install   # Install a package (extension/skill/theme)
+tau remove    # Remove an installed package
+tau list      # List installed packages
+tau update    # Update installed packages
 ```
 
 ### Interactive slash commands

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/docs/cli-reference.md

@@ -15,6 +15,7 @@ augment it.
 |--------|-------|-------------|
 | `--version` | `-v` | Print the installed version |
 | `--debug` | `-d` | Enable debug logging |
+| `--startup` | | Print per-phase startup timing diagnostics to stderr (settings, model/LLM, session manager, resources, extensions, agent) |
 | `--cwd PATH` | `-c` | Set the working directory |
 | `--prompt` | `-p` | Run a non-interactive prompt |
 | `--output-format` | `-f` | Non-interactive output: `text` or `json` |
@@ -384,6 +385,24 @@ Manage provider credentials.
 tau auth --help
 ```
 
+### `tau doctor`
+
+Diagnose Tau's configuration, credentials, model setup, extensions, session
+storage, logs, environment, and installed packages. Reports each check as
+pass/warn/fail; exits non-zero if anything failed.
+
+```bash
+tau doctor              # report only
+tau doctor --json       # machine-readable output
+tau doctor --fix        # apply safe, reversible repairs:
+                         #   - refresh expired OAuth tokens
+                         #   - remove dangling extension entries
+                         #   - quarantine corrupt session files into .corrupt/
+```
+
+`--fix` never rewrites `settings.json`/`auth.json` directly, reinstalls
+packages, or touches anything that isn't easily reversible.
+
 Package management is available through `tau install`, `tau remove`,
 `tau list`, and `tau update`.
 

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/docs/installation.md

@@ -154,6 +154,17 @@ This removes the tau command but leaves configuration and session data in `~/.ta
 
 ## Troubleshooting
 
+Run `tau doctor` first — it checks settings/auth file integrity, credential
+status (including whether stored OAuth tokens are still valid), model/provider
+resolution, extensions, session storage, logs, and installed packages in one
+pass. Add `--fix` to auto-repair the safe, reversible cases (expired OAuth
+tokens, dangling extension entries, corrupt session files).
+
+```bash
+tau doctor
+tau doctor --fix
+```
+
 ### No Models Found
 
 Check that your API key is set correctly:

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "tau-coding-agent"
-version = "0.7.7"
+version = "0.7.8"
 description = "A self-extensible agent CLI with terminal UI, multi-provider LLM support, session management with branching, and a plugin system for tools and commands"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -84,4 +84,5 @@ dev = [
     "ruff>=0.4.0",
     "mypy==1.19.1",
     "pyright>=1.1.360",
+    "pytest-asyncio==1.3.0",
 ]

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/builtins/extensions/computer_use/__init__.py

@@ -16,13 +16,20 @@ extensions.list settings.
   }
 
 While the desktop session is open (after a computer action='open' call), a
-compact desktop-state summary — focused/open windows plus the accessible
-interactive elements on screen (see state.py) — is injected ephemerally into
-LLM context at the start of every turn, the same way the todo extension
-re-asserts its live list: via the "context" hook returning
+compact desktop-state summary is injected ephemerally into LLM context at the
+start of every turn, the same way the todo extension re-asserts its live
+list: via the "context" hook returning
 ContextEventResult(ephemeral_messages=[...]). It is never written to session
 history, so it always reflects the current screen rather than a stale
 snapshot from whenever the model last looked.
+
+The `mode` setting (see manifest.json) controls what that summary contains:
+"screenshot" (an image of the screen), "accessibility_tree" (the default —
+focused/open windows plus accessible interactive elements, see state.py), or
+"both". The screenshot is only ever included if the active model actually
+accepts image input (checked fresh every turn against Modality.Image, since
+/model can switch models mid-session); otherwise the accessibility tree is
+used instead so the turn still gets a usable observation.
 """
 
 from __future__ import annotations
@@ -49,14 +56,16 @@ def register(tau: ExtensionAPI) -> None:
 
     tau.register_tool(ComputerTool(desktop))
 
+    mode = config.get("mode", "accessibility_tree")
+
     def _inject_desktop_state(_event, _ctx: ExtensionContext):
-        text = build_state_message(desktop)
-        if text is not None:
-            from tau.hooks.engine import ContextEventResult
-            from tau.message.types import UserMessage
-            message = UserMessage.from_text(text)
-            return ContextEventResult(ephemeral_messages=[message])
-        return ContextEventResult(ephemeral_messages=[])
+        from tau.hooks.engine import ContextEventResult
+        from tau.inference.model.types import Modality
+
+        model = getattr(_ctx.llm, "model", None)
+        supports_image = model is not None and Modality.Image in model.input
+        message = build_state_message(desktop, mode, supports_image)
+        return ContextEventResult(ephemeral_messages=[message] if message is not None else [])
 
     tau.on("context", _inject_desktop_state)
 

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/builtins/extensions/computer_use/macos/desktop/service.py

@@ -73,13 +73,18 @@ class Desktop(DesktopBase):
 
     # -- State / inspection ---------------------------------------------------
 
-    def get_state(self, as_bytes: bool = False) -> DesktopState:
+    def get_state(
+        self,
+        as_bytes: bool = False,
+        use_screenshot: bool = True,
+        use_accessibility: bool = True,
+    ) -> DesktopState:
         active_window = self.get_foreground_window()
         return DesktopState(
             active_window=active_window,
             windows=self.get_windows(),
-            screenshot=self.get_screenshot(as_bytes=as_bytes),
-            tree_state=self._tree.get_state(active_window),
+            screenshot=self.get_screenshot(as_bytes=as_bytes) if use_screenshot else None,
+            tree_state=self._tree.get_state(active_window) if use_accessibility else None,
         )
 
     def get_screen_size(self) -> Size:

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/builtins/extensions/computer_use/manifest.json

@@ -18,6 +18,14 @@
           "type": "bool",
           "default": false,
           "description": "Register the computer tool, letting the agent control the real mouse, keyboard, and application windows on this machine."
+        },
+        {
+          "key": "mode",
+          "label": "Observation Mode",
+          "type": "enum",
+          "values": ["screenshot", "accessibility_tree", "both"],
+          "default": "accessibility_tree",
+          "description": "What desktop state is injected into context each turn: a screenshot image, the accessibility tree of interactive elements, or both."
         }
       ]
     }

tau_coding_agent-0.7.8/tau/builtins/extensions/computer_use/state.py

@@ -0,0 +1,53 @@
+"""Builds the compact desktop-state summary injected ephemerally into LLM
+context each turn while the computer tool's desktop session is open."""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Literal
+
+from tau.message.types import UserMessage
+
+if TYPE_CHECKING:
+    from .types import Desktop
+
+ObservationMode = Literal["screenshot", "accessibility_tree", "both"]
+
+
+def build_state_message(
+    desktop: Desktop, mode: ObservationMode = "accessibility_tree", supports_image: bool = True
+) -> UserMessage | None:
+    """Live desktop summary: focused/open windows, plus a screenshot and/or
+    the accessible interactive elements currently on screen, per `mode`.
+    None while the desktop is closed, so nothing gets injected before
+    action='open' is called.
+
+    `supports_image` gates the screenshot on the active model's actual
+    modality support — set from `Modality.Image in ctx.llm.model.input`,
+    checked fresh every call since the model can change mid-session via
+    /model. When a screenshot was requested but the model can't take images,
+    the accessibility tree is included instead so the turn isn't left with
+    no usable observation at all.
+    """
+    if not desktop.is_open:
+        return None
+
+    wants_screenshot = mode in ("screenshot", "both")
+    wants_accessibility = mode in ("accessibility_tree", "both")
+    use_screenshot = wants_screenshot and supports_image
+    use_accessibility = wants_accessibility or (wants_screenshot and not supports_image)
+    state = desktop.get_state(
+        as_bytes=True, use_screenshot=use_screenshot, use_accessibility=use_accessibility
+    )
+    parts = [state.to_string()]
+
+    if use_accessibility:
+        elements_to_string = getattr(state.tree_state, "interactive_elements_to_string", None)
+        if callable(elements_to_string):
+            elements = elements_to_string()
+            if elements:
+                parts.append(f"Interactive elements (click at the given coords):\n{elements}")
+
+    text = "\n\n".join(parts)
+    if use_screenshot and state.screenshot is not None:
+        return UserMessage.with_images(text, [state.screenshot])
+    return UserMessage.from_text(text)

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/builtins/extensions/computer_use/types.py

@@ -104,8 +104,15 @@ class Desktop(ABC):
     # -- State / inspection ------------------------------------------------
 
     @abstractmethod
-    def get_state(self, as_bytes: bool = False) -> DesktopState:
-        """Return a full snapshot of the current desktop."""
+    def get_state(
+        self,
+        as_bytes: bool = False,
+        use_screenshot: bool = True,
+        use_accessibility: bool = True,
+    ) -> DesktopState:
+        """Return a snapshot of the current desktop. Skips capturing the
+        screenshot/accessibility tree when the corresponding flag is False,
+        so callers that only need one can avoid paying for the other."""
 
     @abstractmethod
     def get_screen_size(self) -> Size:

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/builtins/extensions/computer_use/windows/desktop/service.py

@@ -86,19 +86,28 @@ class Desktop(DesktopBase):
 
     # -- State / inspection ---------------------------------------------------
 
-    def get_state(self, as_bytes: bool = False) -> DesktopState:
+    def get_state(
+        self,
+        as_bytes: bool = False,
+        use_screenshot: bool = True,
+        use_accessibility: bool = True,
+    ) -> DesktopState:
         active_window = self.get_foreground_window()
-        active_handle = active_window.native_handle if active_window else None
-        other_handles = [
-            control.NativeWindowHandle
-            for control in uia.GetRootControl().GetChildren()
-            if control.NativeWindowHandle and control.NativeWindowHandle != active_handle
-        ]
+        screenshot = self.get_screenshot(as_bytes=as_bytes) if use_screenshot else None
+        tree_state = None
+        if use_accessibility:
+            active_handle = active_window.native_handle if active_window else None
+            other_handles = [
+                control.NativeWindowHandle
+                for control in uia.GetRootControl().GetChildren()
+                if control.NativeWindowHandle and control.NativeWindowHandle != active_handle
+            ]
+            tree_state = self._tree.get_state(active_handle, other_handles)
         return DesktopState(
             active_window=active_window,
             windows=self.get_windows(),
-            screenshot=self.get_screenshot(as_bytes=as_bytes),
-            tree_state=self._tree.get_state(active_handle, other_handles),
+            screenshot=screenshot,
+            tree_state=tree_state,
         )
 
     def get_screen_size(self) -> Size:

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/builtins/extensions/footer/model.py

@@ -78,16 +78,18 @@ class ModelBadge(Component):
             self.update_context_from_ctx(ctx)
             return
 
-        cache_read = (
-            0
-            if getattr(usage, "input_tokens_include_cache_read", False)
-            else getattr(usage, "cache_read_tokens", 0) or 0
-        )
+        # When input_tokens already folds in the cache read/write breakdown
+        # (OpenAI Responses/Chat Completions, Gemini), adding them again here
+        # would double-count; only add them for providers that report cache
+        # tokens as fully separate from input_tokens (Anthropic).
+        includes_cache = getattr(usage, "input_tokens_include_cache_read", False)
+        cache_read = 0 if includes_cache else getattr(usage, "cache_read_tokens", 0) or 0
+        cache_write = 0 if includes_cache else getattr(usage, "cache_write_tokens", 0) or 0
         tokens = (
             (getattr(usage, "input_tokens", 0) or 0)
             + (getattr(usage, "output_tokens", 0) or 0)
             + cache_read
-            + (getattr(usage, "cache_write_tokens", 0) or 0)
+            + cache_write
         )
         context = getattr(ctx, "get_context_usage", lambda: None)()
         window = (context or {}).get("context_window") or self._context_window

{tau_coding_agent-0.7.7 → tau_coding_agent-0.7.8}/tau/console/cli.py

@@ -10,7 +10,9 @@ from typing import TYPE_CHECKING
 import click
 
 from tau.console.commands.auth import auth
+from tau.console.commands.doctor import doctor
 from tau.console.commands.packages import install, list_packages, remove
+from tau.console.commands.serve import serve
 from tau.console.commands.update import update
 from tau.settings.paths import get_app_version
 
@@ -43,6 +45,13 @@ def resolve_model(model: str | None, provider: str | None) -> tuple[str | None,
 @click.group(invoke_without_command=True, context_settings={"help_option_names": ["-h", "--help"]})
 @click.option("--version", "-v", is_flag=True, default=False, help="Print version and exit.")
 @click.option("--debug", "-d", is_flag=True, default=False, help="Enable debug logging.")
+@click.option(
+    "--startup",
+    is_flag=True,
+    default=False,
+    help="Print per-phase startup timing diagnostics to stderr "
+    "(settings, model/LLM, session manager, resources, extensions, agent).",
+)
 @click.option("--cwd", "-c", default=None, metavar="PATH", help="Set the working directory.")
 @click.option(
     "--prompt",
@@ -153,6 +162,7 @@ def cli(
     ctx: click.Context,
     version: bool,
     debug: bool,
+    startup: bool,
     cwd: str | None,
     prompt: str | None,
     output_format: str,
@@ -183,6 +193,11 @@ def cli(
     if debug:
         logging.basicConfig(level=logging.DEBUG, format="%(levelname)s %(name)s: %(message)s")
 
+    if startup:
+        from tau.utils import timing
+
+        timing.enable()
+
     if cwd:
         os.chdir(cwd)
 
@@ -279,6 +294,11 @@ async def _start(opts: dict) -> None:
     )
 
     runtime = await Runtime.create(config)
+
+    from tau.utils import timing
+
+    timing.print_report()
+
     if opts.get("session_name"):
         runtime.session_manager.append_session_info(opts["session_name"])
 
@@ -435,9 +455,11 @@ async def _run_json(runtime: Runtime, message: str | None, quiet: bool = False)
 
 
 cli.add_command(auth)
+cli.add_command(doctor)
 cli.add_command(install)
 cli.add_command(remove)
 cli.add_command(update)
+cli.add_command(serve)
 cli.add_command(list_packages, name="list")