fix(ui): harden dogfood task execution

coding-agent/src/coding_agent/server/session_manager.py

@@ -2456,6 +2456,7 @@ async def run_agent(
                 outcome = await adapter.run_turn(prompt)
                 if self._runtime_store is not None:
                     turn_outcome = self._require_turn_outcome(outcome)
+                    turn_status = self._status_from_turn_outcome(turn_outcome)
                     session.tape_id = ctx.tape.tape_id
                     await self._save_runtime_message_snapshot(
                         session,
@@ -2465,13 +2466,30 @@ async def run_agent(
                     await self._finish_runtime_agent_run(
                         session,
                         run_id=run_id,
-                        status=self._status_from_turn_outcome(turn_outcome),
+                        status=turn_status,
                         result=self._result_from_turn_outcome(turn_outcome),
                         error=turn_outcome.error,
                     )
+                    if turn_status == "failed":
+                        session.turn_status = "failed"
+                        reason = turn_outcome.error or turn_outcome.stop_reason.value
+                        session.last_failure_details = f"Agent turn failed: {reason}"
+                    else:
+                        session.last_failure_details = None
                 else:
                     session.tape_id = ctx.tape.tape_id
-                session.last_failure_details = None
+                    if isinstance(outcome, TurnOutcome):
+                        turn_status = self._status_from_turn_outcome(outcome)
+                        if turn_status == "failed":
+                            session.turn_status = "failed"
+                            reason = outcome.error or outcome.stop_reason.value
+                            session.last_failure_details = (
+                                f"Agent turn failed: {reason}"
+                            )
+                        else:
+                            session.last_failure_details = None
+                    else:
+                        session.last_failure_details = None
                 await self._persist_session_async(session)
             except FatalToolExecutionError as exc:
                 if agent_run_created:

coding-agent/src/coding_agent/tools/file_ops.py

@@ -77,7 +77,10 @@ def _resolve_workspace_path(path: str) -> Path:
     try:
         resolved.relative_to(_workspace_root)
     except ValueError as exc:
-        raise ValueError(f"Path is outside workspace root: {path}") from exc
+        raise ValueError(
+            f"Path is outside workspace root: {path}. "
+            f"Use a path under workspace root: {_workspace_root}"
+        ) from exc
 
     return resolved
 
@@ -111,13 +114,19 @@ def resolve_path(path: str) -> Path:
         try:
             resolved.relative_to(root)
         except ValueError as exc:
-            raise ValueError(f"Path is outside workspace root: {path}") from exc
+            raise ValueError(
+                f"Path is outside workspace root: {path}. "
+                f"Use a path under workspace root: {root}"
+            ) from exc
 
         return resolved
 
     @tool(
         name="file_read",
-        description="Read file contents. Returns file text or error message.",
+        description=(
+            "Read file contents under the workspace root. Returns file text or "
+            "error message."
+        ),
     )
     def bound_file_read(path: str) -> str | dict[str, Any]:
         try:
@@ -132,7 +141,10 @@ def bound_file_read(path: str) -> str | dict[str, Any]:
 
     @tool(
         name="file_write",
-        description="Write content to a file. Creates parent directories if needed.",
+        description=(
+            "Write content to a file under the workspace root. Creates parent "
+            "directories if needed."
+        ),
     )
     def bound_file_write(path: str, content: str) -> str:
         try:
@@ -143,7 +155,10 @@ def bound_file_write(path: str, content: str) -> str:
         except Exception as e:
             return f"Error writing {path}: {e}"
 
-    @tool(name="file_replace", description="Replace exact string in a file.")
+    @tool(
+        name="file_replace",
+        description="Replace exact string in a file under the workspace root.",
+    )
     def bound_file_replace(path: str, old: str, new: str) -> str:
         try:
             p = resolve_path(path)
@@ -158,7 +173,10 @@ def bound_file_replace(path: str, old: str, new: str) -> str:
         except Exception as e:
             return f"Error: {e}"
 
-    @tool(name="glob_files", description="Search for files matching a glob pattern.")
+    @tool(
+        name="glob_files",
+        description="Search under the workspace root for files matching a glob pattern.",
+    )
     def bound_glob_files(pattern: str, directory: str = ".") -> str:
         try:
             base = resolve_path(directory)
@@ -169,7 +187,10 @@ def bound_glob_files(pattern: str, directory: str = ".") -> str:
         except Exception as e:
             return f"Error: {e}"
 
-    @tool(name="grep_search", description="Search file contents for a regex pattern.")
+    @tool(
+        name="grep_search",
+        description="Search file contents under the workspace root for a regex pattern.",
+    )
     def bound_grep_search(
         pattern: str, directory: str = ".", include: str = ""
     ) -> str | dict[str, Any]:
@@ -204,27 +225,37 @@ def bound_grep_search(
     return tools
 
 
-@tool(description="Read file contents. Returns file text or error message.")
+@tool(
+    description=(
+        "Read file contents under the workspace root. Returns file text or error "
+        "message."
+    )
+)
 def file_read(path: str) -> str | dict[str, Any]:
     return build_file_tools(_workspace_root)[0](path)
 
 
-@tool(description="Write content to a file. Creates parent directories if needed.")
+@tool(
+    description=(
+        "Write content to a file under the workspace root. Creates parent "
+        "directories if needed."
+    )
+)
 def file_write(path: str, content: str) -> str:
     return build_file_tools(_workspace_root)[1](path, content)
 
 
-@tool(description="Replace exact string in a file.")
+@tool(description="Replace exact string in a file under the workspace root.")
 def file_replace(path: str, old: str, new: str) -> str:
     return build_file_tools(_workspace_root)[2](path, old, new)
 
 
-@tool(description="Search for files matching a glob pattern.")
+@tool(description="Search under the workspace root for files matching a glob pattern.")
 def glob_files(pattern: str, directory: str = ".") -> str:
     return build_file_tools(_workspace_root)[3](pattern, directory)
 
 
-@tool(description="Search file contents for a regex pattern.")
+@tool(description="Search file contents under the workspace root for a regex pattern.")
 def grep_search(
     pattern: str, directory: str = ".", include: str = ""
 ) -> str | dict[str, Any]:

coding-agent/src/coding_agent/tools/shell.py

@@ -182,6 +182,26 @@ def _optional_int(value: object | None) -> int | None:
     return int(value)
 
 
+def _positive_int(value: object, *, name: str) -> int:
+    if isinstance(value, bool):
+        raise ValueError(f"{name} must be a positive integer")
+    if isinstance(value, int):
+        parsed = value
+    elif isinstance(value, str):
+        stripped = value.strip()
+        if not stripped:
+            raise ValueError(f"{name} must be a positive integer")
+        try:
+            parsed = int(stripped)
+        except ValueError as exc:
+            raise ValueError(f"{name} must be a positive integer") from exc
+    else:
+        raise ValueError(f"{name} must be a positive integer")
+    if parsed <= 0:
+        raise ValueError(f"{name} must be a positive integer")
+    return parsed
+
+
 def _sandbox_request(
     *, args: list[str], cwd: str | None, env: dict[str, str] | None, timeout: int
 ) -> object:
@@ -205,7 +225,8 @@ def _validated_execution_cwd(
         resolved_cwd.relative_to(workspace_root)
     except ValueError as exc:
         raise ValueError(
-            f"Working directory is outside sandbox workspace: {resolved_cwd}"
+            "Working directory is outside sandbox workspace: "
+            f"{resolved_cwd}. Use a path under workspace root: {workspace_root}"
         ) from exc
     return str(resolved_cwd)
 
@@ -218,11 +239,18 @@ def _validate_no_path_escape(args: list[str], workspace_root: Path) -> None:
                 candidate.relative_to(workspace_root)
             except ValueError as exc:
                 raise ValueError(
-                    f"Path is outside sandbox workspace: {candidate}"
+                    "Path is outside sandbox workspace: "
+                    f"{candidate}. Use a path under workspace root: {workspace_root}"
                 ) from exc
 
 
-@tool(description="Run a shell command and return stdout/stderr.")
+@tool(
+    description=(
+        "Run one simple command and return stdout/stderr. Use separate bash_run "
+        "calls instead of &&, ||, pipes, redirects, heredocs, semicolons, or "
+        "backgrounding. Paths must stay under the workspace root."
+    )
+)
 def bash_run(
     command: str,
     timeout: int = 120,
@@ -231,6 +259,7 @@ def bash_run(
     __pipeline_ctx__: object | None = None,
 ) -> str | dict[str, str | int]:
     try:
+        timeout_seconds = _positive_int(timeout, name="timeout")
         shell_config = _pipeline_shell_config(__pipeline_ctx__)
         mode_value = _sandbox_mode(shell_config)
         execution_cwd = _validated_execution_cwd(
@@ -251,7 +280,9 @@ def bash_run(
                     Path(changed_dir).relative_to(workspace_root)
                 except ValueError as exc:
                     raise ValueError(
-                        f"Directory is outside sandbox workspace: {changed_dir}"
+                        "Directory is outside sandbox workspace: "
+                        f"{changed_dir}. Use a path under workspace root: "
+                        f"{workspace_root}"
                     ) from exc
             return f"Changed directory to {changed_dir}"
 
@@ -273,7 +304,7 @@ def bash_run(
                 shell=False,
                 capture_output=True,
                 text=True,
-                timeout=timeout,
+                timeout=timeout_seconds,
                 cwd=execution_cwd,
                 env=_build_env(env),
             )
@@ -287,7 +318,7 @@ def bash_run(
                     args=args,
                     cwd=execution_cwd,
                     env=env,
-                    timeout=timeout,
+                    timeout=timeout_seconds,
                 )
             )
         if _structured_results_enabled():
@@ -301,7 +332,7 @@ def bash_run(
             output += f"\nExit code: {result.returncode}"
         return output.strip() or "(no output)"
     except subprocess.TimeoutExpired:
-        return f"Error: command timed out after {timeout}s"
+        return f"Error: command timed out after {timeout_seconds}s"
     except Exception as e:
         return f"Error: {e}"
 

coding-agent/tests/coding_agent/test_cli_pipeline.py

@@ -79,9 +79,7 @@ async def test_tui_run_uses_pipeline_adapter(self):
                 "coding_agent.__main__.create_agent",
                 return_value=(mock_pipeline, mock_ctx),
             ) as p_create,
-            patch(
-                "coding_agent.__main__.PipelineAdapter", mock_adapter_cls
-            ) as p_adapter,
+            patch("coding_agent.__main__.PipelineAdapter", mock_adapter_cls),
             patch("coding_agent.__main__.CodingAgentTUI", mock_tui_cls),
         ):
             from coding_agent.__main__ import _run_with_tui
@@ -606,16 +604,9 @@ async def fake_run_turn(msg: str) -> TurnOutcome:
             )
             return _ok_outcome()
 
-        with patch("coding_agent.cli.repl.CodingAgentTUI") as mock_tui_cls:
-            mock_tui = MagicMock()
-            mock_tui.__enter__ = MagicMock(return_value=mock_tui)
-            mock_tui.__exit__ = MagicMock(return_value=False)
-            mock_tui.consumer = MagicMock()
-            mock_tui_cls.return_value = mock_tui
-
-            with patch.object(adapter, "run_turn", side_effect=fake_run_turn):
-                await session._process_message("first question")
-                await session._process_message("second question")
+        with patch.object(adapter, "run_turn", side_effect=fake_run_turn):
+            await session._process_message("first question")
+            await session._process_message("second question")
 
         assert call_messages == ["first question", "second question"]
         assert session._pipeline_adapter is adapter
@@ -697,18 +688,9 @@ async def side_effect_run_turn(msg: str) -> TurnOutcome:
             patch("coding_agent.cli.commands.print_pt"),
             patch("coding_agent.cli.commands.print_html"),
         ):
-            with patch("coding_agent.cli.repl.CodingAgentTUI") as mock_tui_cls:
-                mock_tui = MagicMock()
-                mock_tui.__enter__ = MagicMock(return_value=mock_tui)
-                mock_tui.__exit__ = MagicMock(return_value=False)
-                mock_tui.consumer = MagicMock()
-                mock_tui_cls.return_value = mock_tui
-
-                with patch.object(
-                    adapter, "run_turn", side_effect=side_effect_run_turn
-                ):
-                    await session._process_message("bad input")
-                    await session._process_message("good input")
+            with patch.object(adapter, "run_turn", side_effect=side_effect_run_turn):
+                await session._process_message("bad input")
+                await session._process_message("good input")
 
         assert len(turn_results) == 2
         assert turn_results[0].stop_reason == StopReason.ERROR
@@ -809,8 +791,8 @@ async def test_headless_pipeline_no_final_message_omits_echo(self, capsys):
         # Result summary line present
         assert "--- Result" in captured.out
         # No additional content line after the result summary
-        lines = [l for l in captured.out.strip().split("\n") if l.strip()]
-        result_lines = [l for l in lines if "--- Result" in l]
+        lines = [line for line in captured.out.strip().split("\n") if line.strip()]
+        result_lines = [line for line in lines if "--- Result" in line]
         assert len(result_lines) == 1
 
 
@@ -979,7 +961,7 @@ def test_repl_sets_ask_user_handler_on_directive_executor(self, mock_create_agen
         from coding_agent.cli.repl import InteractiveSession
 
         config = _make_repl_config()
-        session = InteractiveSession(config)
+        _ = InteractiveSession(config)
 
         assert mock_pipeline._directive_executor._ask_user is not None
         assert callable(mock_pipeline._directive_executor._ask_user)
@@ -1000,7 +982,7 @@ async def test_repl_approval_prompt_approved(self, mock_create_agent, monkeypatc
         from coding_agent.cli.repl import InteractiveSession
 
         config = _make_repl_config()
-        session = InteractiveSession(config)
+        _ = InteractiveSession(config)
 
         assert executor._ask_user is not None
 
@@ -1026,7 +1008,7 @@ async def test_repl_approval_prompt_denied(self, mock_create_agent, monkeypatch)
         from coding_agent.cli.repl import InteractiveSession
 
         config = _make_repl_config()
-        session = InteractiveSession(config)
+        _ = InteractiveSession(config)
 
         monkeypatch.setattr("builtins.input", lambda prompt="": "n")
 
@@ -1051,7 +1033,7 @@ async def test_repl_approval_prompt_empty_defaults_no(
         from coding_agent.cli.repl import InteractiveSession
 
         config = _make_repl_config()
-        session = InteractiveSession(config)
+        _ = InteractiveSession(config)
 
         monkeypatch.setattr("builtins.input", lambda prompt="": "")
 
@@ -1075,7 +1057,7 @@ async def test_repl_approval_yes_variants(self, mock_create_agent, monkeypatch):
         from coding_agent.cli.repl import InteractiveSession
 
         config = _make_repl_config()
-        session = InteractiveSession(config)
+        _ = InteractiveSession(config)
 
         for variant in ("y", "Y", "yes", "YES", "Yes"):
             monkeypatch.setattr("builtins.input", lambda prompt="", v=variant: v)