wafer-cli 0.2.19__tar.gz → 0.2.21__tar.gz

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: wafer-cli
-Version: 0.2.19
+Version: 0.2.21
 Summary: CLI tool for running commands on remote GPUs and GPU kernel optimization agent
 Requires-Python: >=3.11
 Requires-Dist: typer>=0.12.0

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "wafer-cli"
-version = "0.2.19"
+version = "0.2.21"
 description = "CLI tool for running commands on remote GPUs and GPU kernel optimization agent"
 requires-python = ">=3.11"
 dependencies = [

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/tests/test_cli_coverage.py

@@ -719,3 +719,17 @@ class TestWorkspacesExecFlagPassthrough:
             "workspaces", "exec", "test-ws", "--", "cmd", "--output=/tmp/out"
         ])
         assert "no such option" not in result.output.lower()
+
+
+class TestAgentNoSandboxOption:
+    """Test --no-sandbox option in wafer agent command."""
+
+    def test_agent_no_sandbox_option_exists(self) -> None:
+        """Test that --no-sandbox option is accepted by wafer agent command."""
+        result = runner.invoke(app, ["agent", "--help"])
+        assert result.exit_code == 0
+        # Strip ANSI escape codes before checking (help output may contain color codes)
+        ansi_escape = re.compile(r'\x1b\[[0-9;]*m')
+        clean_output = ansi_escape.sub('', result.stdout)
+        assert "--no-sandbox" in clean_output
+        assert "liability" in clean_output.lower()  # Warning text should be in help

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/tests/test_wevin_cli.py

@@ -634,35 +634,114 @@ def test_streaming_frontend_session_start_state_without_session_id():
 
 def test_streaming_frontend_session_start_resumed_then_new():
     """Test session_start emission when resuming but states have different session_id.
-    
+
     Edge case: --resume used but states return different session_id (should use states one).
     """
     import trio
 
     from wafer.wevin_cli import StreamingChunkFrontend
-    
+
     async def run_test() -> None:
         # Start with resumed session_id
         frontend = StreamingChunkFrontend(
             session_id="resumed-session-123",
             model="claude-sonnet-4.5"
         )
-        
+
         emitted_events = []
 
         def mock_emit(obj) -> None:
             emitted_events.append(obj)
-        
+
         frontend._emit = mock_emit
-        
+
         # start() emits session_start for resumed session
         await frontend.start()
         assert len(emitted_events) == 1
         assert emitted_events[0]["session_id"] == "resumed-session-123"
-        
+
         # If states have different session_id (shouldn't happen, but handle gracefully)
         # The logic in main() checks `if first_session_id and not session_id`
         # So if session_id was set, it won't emit again
         # This is correct behavior - use the one from --resume
-    
+
     trio.run(run_test)
+
+
+# =============================================================================
+# --no-sandbox flag tests
+# =============================================================================
+
+
+def test_no_sandbox_parameter_accepted():
+    """Test that no_sandbox parameter exists in wevin_main signature."""
+    import inspect
+
+    from wafer.wevin_cli import main as wevin_main
+
+    sig = inspect.signature(wevin_main)
+    params = sig.parameters
+
+    # Verify parameter exists
+    assert 'no_sandbox' in params
+
+    # Verify type and default
+    assert str(params['no_sandbox'].annotation) in ('bool', "<class 'bool'>")
+    assert params['no_sandbox'].default is False
+
+
+def test_build_environment_accepts_no_sandbox():
+    """Test that _build_environment accepts no_sandbox parameter."""
+    import inspect
+
+    from wafer.wevin_cli import _build_environment
+
+    sig = inspect.signature(_build_environment)
+    params = sig.parameters
+
+    assert 'no_sandbox' in params
+    assert params['no_sandbox'].default is False
+
+
+def test_build_environment_with_no_sandbox_false():
+    """Test _build_environment creates env with sandbox ENABLED when no_sandbox=False."""
+    from wafer_core.rollouts.templates import TemplateConfig
+    from wafer_core.sandbox import SandboxMode
+
+    from wafer.wevin_cli import _build_environment
+
+    tpl = TemplateConfig(
+        name="test",
+        description="Test template",
+        system_prompt="Test",
+        tools=["read"],
+    )
+
+    # This will raise RuntimeError if sandbox is unavailable on this system
+    # That's expected - we're testing that sandbox is ENABLED by default
+    try:
+        env = _build_environment(tpl, None, None, no_sandbox=False)
+        # If we get here, sandbox is available - verify it's enabled
+        assert env.sandbox_mode == SandboxMode.ENABLED
+    except RuntimeError as e:
+        # Sandbox unavailable - that's OK, the error proves ENABLED is set
+        assert "sandboxing is not available" in str(e)
+
+
+def test_build_environment_with_no_sandbox_true():
+    """Test _build_environment creates env with sandbox DISABLED when no_sandbox=True."""
+    from wafer_core.rollouts.templates import TemplateConfig
+    from wafer_core.sandbox import SandboxMode
+
+    from wafer.wevin_cli import _build_environment
+
+    tpl = TemplateConfig(
+        name="test",
+        description="Test template",
+        system_prompt="Test",
+        tools=["read"],
+    )
+
+    # This should NOT raise - sandbox is disabled
+    env = _build_environment(tpl, None, None, no_sandbox=True)
+    assert env.sandbox_mode == SandboxMode.DISABLED

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/wafer/cli.py

@@ -1,6 +1,8 @@
-# ruff: noqa: PLR0913
+# ruff: noqa: PLR0913, E402
 # PLR0913 (too many arguments) is suppressed because Typer CLI commands
 # naturally have many parameters - each --flag becomes a function argument.
+# E402 (module level import not at top) is suppressed because we intentionally
+# load .env files before importing other modules that may read env vars.
 """Wafer CLI - GPU development toolkit for LLM coding agents.
 
 Core commands:
@@ -27,6 +29,12 @@ from pathlib import Path
 
 import trio
 import typer
+from dotenv import load_dotenv
+
+# Auto-load .env from current directory and ~/.wafer/.env
+# This runs at import time so env vars are available before any config is accessed
+load_dotenv()  # cwd/.env
+load_dotenv(Path.home() / ".wafer" / ".env")  # ~/.wafer/.env
 
 from .config import WaferConfig, WaferEnvironment
 from .inference import infer_upload_files, resolve_environment
@@ -42,6 +50,7 @@ from .problems import (
 app = typer.Typer(
     help="GPU development toolkit for LLM coding agents",
     no_args_is_help=True,
+    pretty_exceptions_show_locals=False,  # Don't dump local vars (makes tracebacks huge)
 )
 
 # =============================================================================
@@ -58,11 +67,11 @@ def _show_version() -> None:
     """Show CLI version and environment, then exit."""
     from .analytics import _get_cli_version
     from .global_config import load_global_config
-    
+
     version = _get_cli_version()
     config = load_global_config()
     environment = config.environment
-    
+
     typer.echo(f"wafer-cli {version} ({environment})")
     raise typer.Exit()
 
@@ -110,7 +119,7 @@ def main_callback(
     if version:
         _show_version()
         return
-    
+
     global _command_start_time, _command_outcome
     _command_start_time = time.time()
     _command_outcome = "success"  # Default to success, mark failure on exceptions
@@ -121,6 +130,7 @@ def main_callback(
     analytics.init_analytics()
 
     # Install exception hook to catch SystemExit and mark failures
+    # Also prints error message FIRST so it's visible even when traceback is truncated
     original_excepthook = sys.excepthook
 
     def custom_excepthook(
@@ -136,7 +146,9 @@ def main_callback(
                 _command_outcome = "failure"
         else:
             _command_outcome = "failure"
-        # Call original excepthook
+            # Print error summary FIRST (before traceback) so it's visible even if truncated
+            print(f"\n\033[1;31m>>> ERROR: {exc_type.__name__}: {exc_value}\033[0m\n", file=sys.stderr)
+        # Call original excepthook (prints the full traceback)
         original_excepthook(exc_type, exc_value, exc_traceback)
 
     sys.excepthook = custom_excepthook
@@ -591,7 +603,7 @@ app.add_typer(provider_auth_app, name="auth")
 def provider_auth_login(
     provider: str = typer.Argument(
         ...,
-        help="Provider name: runpod, digitalocean, or modal",
+        help="Provider name: runpod, digitalocean, modal, anthropic, or openai",
     ),
     api_key: str | None = typer.Option(
         None,
@@ -600,15 +612,16 @@ def provider_auth_login(
         help="API key (if not provided, reads from stdin)",
     ),
 ) -> None:
-    """Save API key for a cloud GPU provider.
+    """Save API key for a provider.
 
     Stores the key in ~/.wafer/auth.json. Environment variables
-    (e.g., WAFER_RUNPOD_API_KEY) take precedence over stored keys.
+    (e.g., ANTHROPIC_API_KEY) take precedence over stored keys.
 
     Examples:
+        wafer auth login anthropic --api-key sk-ant-xxx
         wafer auth login runpod --api-key rp_xxx
-        wafer auth login digitalocean --api-key dop_v1_xxx
-        echo $API_KEY | wafer auth login runpod
+        wafer auth login openai --api-key sk-xxx
+        echo $API_KEY | wafer auth login anthropic
     """
     import sys
 
@@ -642,7 +655,7 @@ def provider_auth_login(
 def provider_auth_logout(
     provider: str = typer.Argument(
         ...,
-        help="Provider name: runpod, digitalocean, or modal",
+        help="Provider name: runpod, digitalocean, modal, anthropic, or openai",
     ),
 ) -> None:
     """Remove stored API key for a cloud GPU provider.
@@ -1327,6 +1340,11 @@ def agent(  # noqa: PLR0913
         "-c",
         help="Documentation corpus to use (cuda, cutlass, hip, amd). Must be downloaded first.",
     ),
+    no_sandbox: bool = typer.Option(
+        False,
+        "--no-sandbox",
+        help="Disable OS-level sandboxing (YOU accept liability for any damage caused by the agent)",
+    ),
 ) -> None:
     """AI assistant for GPU kernel development.
 
@@ -1408,6 +1426,13 @@ def agent(  # noqa: PLR0913
                 raise typer.Exit(1) from None
         corpus_path = str(path)
 
+    # Warn user about sandbox disabled
+    if no_sandbox:
+        print(
+            "Warning: Sandbox disabled. You accept liability for any damage caused by the agent.",
+            file=sys.stderr,
+        )
+
     wevin_main(
         prompt=actual_prompt,
         interactive=use_tui,
@@ -1425,6 +1450,7 @@ def agent(  # noqa: PLR0913
         template=template,
         template_args=parsed_template_args,
         corpus_path=corpus_path,
+        no_sandbox=no_sandbox,
     )
 
 
@@ -1455,6 +1481,7 @@ def _make_agent_alias(name: str, doc: str) -> None:
         template: str | None = typer.Option(None, "--template", "-t"),
         template_args: list[str] | None = typer.Option(None, "--args"),
         corpus: str | None = typer.Option(None, "--corpus"),
+        no_sandbox: bool = typer.Option(False, "--no-sandbox"),
     ) -> None:
         agent(
             prompt=prompt,
@@ -1474,6 +1501,7 @@ def _make_agent_alias(name: str, doc: str) -> None:
             template=template,
             template_args=template_args,
             corpus=corpus,
+            no_sandbox=no_sandbox,
         )
 
     alias_cmd.__doc__ = doc
@@ -4391,9 +4419,13 @@ def workspaces_list(
 @workspaces_app.command("create")
 def workspaces_create(
     name: str = typer.Argument(..., help="Workspace name"),
-    gpu_type: str = typer.Option("B200", "--gpu", "-g", help="GPU type: MI300X (AMD) or B200 (NVIDIA, default)"),
+    gpu_type: str = typer.Option(
+        "B200", "--gpu", "-g", help="GPU type: MI300X (AMD) or B200 (NVIDIA, default)"
+    ),
     image: str | None = typer.Option(None, "--image", "-i", help="Docker image (optional)"),
-    wait: bool = typer.Option(False, "--wait", "-w", help="Wait for provisioning and show SSH credentials"),
+    wait: bool = typer.Option(
+        False, "--wait", "-w", help="Wait for provisioning and show SSH credentials"
+    ),
     json_output: bool = typer.Option(False, "--json", "-j", help="Output as JSON"),
 ) -> None:
     """Create a new workspace.
@@ -4702,19 +4734,25 @@ def workspaces_ssh(
     ssh_host = ws.get("ssh_host")
     ssh_port = ws.get("ssh_port")
     ssh_user = ws.get("ssh_user")
-    
+
     if not ssh_host or not ssh_port or not ssh_user:
         typer.echo("Error: Workspace not ready. Wait a few seconds and retry.", err=True)
         raise typer.Exit(1)
 
     # Connect via SSH
-    os.execvp("ssh", [
+    os.execvp(
         "ssh",
-        "-p", str(ssh_port),
-        "-o", "StrictHostKeyChecking=no",
-        "-o", "UserKnownHostsFile=/dev/null",
-        f"{ssh_user}@{ssh_host}",
-    ])
+        [
+            "ssh",
+            "-p",
+            str(ssh_port),
+            "-o",
+            "StrictHostKeyChecking=no",
+            "-o",
+            "UserKnownHostsFile=/dev/null",
+            f"{ssh_user}@{ssh_host}",
+        ],
+    )
 
 
 @workspaces_app.command("sync")

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/wafer/templates/optimize_kernel.py

@@ -68,4 +68,6 @@ IMPORTANT: Always verify correctness with wafer evaluate before claiming success
         "kernel": "./kernel.cu",
         "target": "H100",
     },
+    # Enable skill discovery (agent can load wafer-guide, etc.)
+    include_skills=True,
 )

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/wafer/wevin_cli.py

@@ -266,18 +266,27 @@ def _build_environment(
     tpl: TemplateConfig,
     tools_override: list[str] | None,
     corpus_path: str | None,
+    no_sandbox: bool = False,
 ) -> Environment:
     """Build a CodingEnvironment from template config."""
     from wafer_core.environments.coding import CodingEnvironment
     from wafer_core.rollouts.templates import DANGEROUS_BASH_COMMANDS
+    from wafer_core.sandbox import SandboxMode
 
     working_dir = Path(corpus_path) if corpus_path else Path.cwd()
-    resolved_tools = tools_override or tpl.tools
+    resolved_tools = list(tools_override or tpl.tools)
+
+    # Add skill tool if skills are enabled
+    if tpl.include_skills and "skill" not in resolved_tools:
+        resolved_tools.append("skill")
+
+    sandbox_mode = SandboxMode.DISABLED if no_sandbox else SandboxMode.ENABLED
     env: Environment = CodingEnvironment(
         working_dir=working_dir,
         enabled_tools=resolved_tools,
         bash_allowlist=tpl.bash_allowlist,
         bash_denylist=DANGEROUS_BASH_COMMANDS,
+        sandbox_mode=sandbox_mode,
     )  # type: ignore[assignment]
     return env
 
@@ -362,6 +371,7 @@ def main(  # noqa: PLR0913, PLR0915
     list_sessions: bool = False,
     get_session: str | None = None,
     json_output: bool = False,
+    no_sandbox: bool = False,
 ) -> None:
     """Run wevin agent in-process via rollouts."""
     from dataclasses import asdict
@@ -373,6 +383,7 @@ def main(  # noqa: PLR0913, PLR0915
 
     # Handle --get-session: load session by ID and print
     if get_session:
+
         async def _get_session() -> None:
             try:
                 session, err = await session_store.get(get_session)
@@ -393,16 +404,18 @@ def main(  # noqa: PLR0913, PLR0915
                         error_msg = f"Failed to serialize messages: {e}"
                         print(json.dumps({"error": error_msg}))
                         sys.exit(1)
-                    
-                    print(json.dumps({
-                        "session_id": session.session_id,
-                        "status": session.status.value,
-                        "model": session.endpoint.model if session.endpoint else None,
-                        "created_at": session.created_at,
-                        "updated_at": session.updated_at,
-                        "messages": messages_data,
-                        "tags": session.tags,
-                    }))
+
+                    print(
+                        json.dumps({
+                            "session_id": session.session_id,
+                            "status": session.status.value,
+                            "model": session.endpoint.model if session.endpoint else None,
+                            "created_at": session.created_at,
+                            "updated_at": session.updated_at,
+                            "messages": messages_data,
+                            "tags": session.tags,
+                        })
+                    )
                 else:
                     print(f"Session: {session.session_id}")
                     print(f"Status: {session.status.value}")
@@ -490,7 +503,7 @@ def main(  # noqa: PLR0913, PLR0915
             print(f"Error loading template: {err}", file=sys.stderr)
             sys.exit(1)
         tpl = loaded_template
-        system_prompt = tpl.interpolate_prompt(template_args or {})
+        base_system_prompt = tpl.interpolate_prompt(template_args or {})
         # Show template info when starting without a prompt
         if not prompt and tpl.description:
             print(f"Template: {tpl.name}", file=sys.stderr)
@@ -498,14 +511,27 @@ def main(  # noqa: PLR0913, PLR0915
             print(file=sys.stderr)
     else:
         tpl = _get_default_template()
-        system_prompt = tpl.system_prompt
+        base_system_prompt = tpl.system_prompt
+
+    # Append skill metadata if skills are enabled
+    if tpl.include_skills:
+        from wafer_core.rollouts.skills import discover_skills, format_skill_metadata_for_prompt
+
+        skill_metadata = discover_skills()
+        if skill_metadata:
+            skill_section = format_skill_metadata_for_prompt(skill_metadata)
+            system_prompt = base_system_prompt + "\n\n" + skill_section
+        else:
+            system_prompt = base_system_prompt
+    else:
+        system_prompt = base_system_prompt
 
     # CLI args override template values
     resolved_single_turn = single_turn if single_turn is not None else tpl.single_turn
 
     # Build endpoint and environment
     endpoint = _build_endpoint(tpl, model, api_base, api_key)
-    environment = _build_environment(tpl, tools, corpus_path)
+    environment = _build_environment(tpl, tools, corpus_path, no_sandbox)
 
     # Session store
     session_store = FileSessionStore()
@@ -545,7 +571,7 @@ def main(  # noqa: PLR0913, PLR0915
             else:
                 if json_output:
                     # Emit session_start if we have a session_id (from --resume)
-                    model_name = endpoint.model if hasattr(endpoint, 'model') else None
+                    model_name = endpoint.model if hasattr(endpoint, "model") else None
                     frontend = StreamingChunkFrontend(session_id=session_id, model=model_name)
                 else:
                     frontend = NoneFrontend(show_tool_calls=True, show_thinking=False)
@@ -560,9 +586,11 @@ def main(  # noqa: PLR0913, PLR0915
                 # Emit session_start for new sessions (if session_id was None and we got one)
                 # Check first state to emit as early as possible
                 if json_output and isinstance(frontend, StreamingChunkFrontend):
-                    first_session_id = states[0].session_id if states and states[0].session_id else None
+                    first_session_id = (
+                        states[0].session_id if states and states[0].session_id else None
+                    )
                     if first_session_id and not session_id:  # New session created
-                        model_name = endpoint.model if hasattr(endpoint, 'model') else None
+                        model_name = endpoint.model if hasattr(endpoint, "model") else None
                         frontend.emit_session_start(first_session_id, model_name)
                 # Print resume command with full wafer agent prefix
                 if states and states[-1].session_id:

{wafer_cli-0.2.19 → wafer_cli-0.2.21}/wafer_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: wafer-cli
-Version: 0.2.19
+Version: 0.2.21
 Summary: CLI tool for running commands on remote GPUs and GPU kernel optimization agent
 Requires-Python: >=3.11
 Requires-Dist: typer>=0.12.0