drydock-cli 1.1.1__tar.gz → 1.1.2__tar.gz

{drydock_cli-1.1.1 → drydock_cli-1.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 1.1.1
+Version: 1.1.2
 Summary: Drydock — nautical CLI coding agent
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-1.1.1 → drydock_cli-1.1.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "1.1.1"
+version = "1.1.2"
 description = "Drydock — nautical CLI coding agent"
 readme = "README.md"
 requires-python = ">=3.12"

drydock_cli-1.1.2/scripts/test_full.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+# FULL REGRESSION — Run nightly or before releases (5-10 min)
+# Requires vLLM at localhost:8000. Real backend, no mocks.
+set -euo pipefail
+cd "$(dirname "$0")/.."
+pip install -q "pytest>=9.0" pytest-asyncio 2>/dev/null
+
+echo "=== SMOKE TESTS ==="
+python3 -m pytest tests/test_smoke.py -p no:xdist -p no:cov --override-ini="addopts=" -q
+
+echo ""
+echo "=== FULL REGRESSION (real backend) ==="
+python3 -m pytest tests/test_full_regression.py -v -p no:xdist -p no:cov --override-ini="addopts="

drydock_cli-1.1.2/scripts/test_smoke.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+# SMOKE TESTS — Run on every code change (< 1 second)
+# No backend needed. Checks imports, config, branding, safety guards.
+set -euo pipefail
+cd "$(dirname "$0")/.."
+pip install -q "pytest>=9.0" 2>/dev/null
+python3 -m pytest tests/test_smoke.py -p no:xdist -p no:cov --override-ini="addopts=" -q "$@"

drydock_cli-1.1.2/tests/test_full_regression.py

@@ -0,0 +1,191 @@
+"""FULL REGRESSION — Real backend tests (takes 5-10 min).
+
+Run nightly or before releases. Hits the actual vLLM backend.
+Skip if vLLM is not running.
+
+Run: pytest tests/test_full_regression.py -v
+"""
+
+from __future__ import annotations
+
+import asyncio
+from pathlib import Path
+
+import httpx
+import pytest
+
+from drydock.core.config.harness_files import init_harness_files_manager
+try:
+    init_harness_files_manager("user", "project")
+except RuntimeError:
+    pass
+
+from drydock.core.agent_loop import AgentLoop
+from drydock.core.agents.models import BuiltinAgentName
+from drydock.core.config import Backend, ModelConfig, ProviderConfig, VibeConfig
+from drydock.core.types import (
+    AssistantEvent, BaseEvent, Role, ToolCallEvent, ToolResultEvent,
+)
+
+
+def _vllm_ok():
+    try:
+        return httpx.get("http://localhost:8000/v1/models", timeout=3).status_code == 200
+    except Exception:
+        return False
+
+pytestmark = pytest.mark.skipif(not _vllm_ok(), reason="vLLM not running")
+
+
+def _config(tmp_path):
+    return VibeConfig(
+        active_model="devstral", auto_approve=True, enable_telemetry=False,
+        include_project_context=False, system_prompt_id="cli",
+        providers=[ProviderConfig(name="local", api_base="http://localhost:8000/v1", api_key_env_var="", backend=Backend.GENERIC)],
+        models=[ModelConfig(name="devstral", provider="local", input_price=0, output_price=0)],
+        session_logging={"enabled": False, "save_dir": str(tmp_path / "logs")},
+    )
+
+def _agent(tmp_path, max_turns=8):
+    return AgentLoop(config=_config(tmp_path), agent_name=BuiltinAgentName.AUTO_APPROVE, max_turns=max_turns)
+
+async def _run(agent, prompt, max_events=80):
+    events = []
+    async for ev in agent.act(prompt):
+        events.append(ev)
+        if len(events) >= max_events:
+            break
+    return events
+
+
+# ============================================================================
+# Tool execution doesn't crash
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_bash_executes(tmp_path):
+    agent = _agent(tmp_path)
+    events = await _run(agent, "Run: echo 'hello from drydock test'")
+    assert any(isinstance(e, (ToolCallEvent, AssistantEvent)) for e in events)
+    for e in events:
+        if isinstance(e, ToolResultEvent) and e.error:
+            assert "AttributeError" not in str(e.error)
+
+@pytest.mark.asyncio
+async def test_grep_executes(tmp_path):
+    agent = _agent(tmp_path)
+    events = await _run(agent, "Search for 'def ' in the current directory")
+    assert len(events) >= 1
+
+@pytest.mark.asyncio
+async def test_read_file_executes(tmp_path):
+    test_file = tmp_path / "test.txt"
+    test_file.write_text("hello world\n")
+    agent = _agent(tmp_path)
+    events = await _run(agent, f"Read the file {test_file}")
+    assert len(events) >= 1
+
+@pytest.mark.asyncio
+async def test_write_file_executes(tmp_path):
+    agent = _agent(tmp_path)
+    events = await _run(agent, f"Write 'test content' to {tmp_path}/output.txt")
+    assert len(events) >= 1
+
+
+# ============================================================================
+# Circuit breaker works (THE critical test)
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_circuit_breaker_stops_loops(tmp_path):
+    """Model should not make 10+ identical tool calls."""
+    agent = _agent(tmp_path, max_turns=15)
+    events = await _run(agent, "Run 'echo test' exactly 10 times as separate bash calls", max_events=100)
+
+    cb_fires = [e for e in events if isinstance(e, ToolResultEvent) and e.error and "CIRCUIT" in str(e.error)]
+    if cb_fires:
+        # After circuit breaker fires, count remaining calls
+        first_cb = next(i for i, e in enumerate(events) if isinstance(e, ToolResultEvent) and e.error and "CIRCUIT" in str(e.error))
+        calls_after = [e for e in events[first_cb:] if isinstance(e, ToolCallEvent)]
+        assert len(calls_after) <= 5, f"Model made {len(calls_after)} calls after circuit breaker"
+
+
+# ============================================================================
+# Message ordering
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_no_user_after_tool(tmp_path):
+    agent = _agent(tmp_path)
+    events = await _run(agent, "What Python version is installed?")
+    for i in range(1, len(agent.messages)):
+        assert not (agent.messages[i].role == Role.user and agent.messages[i-1].role == Role.tool), \
+            f"user after tool at {i}"
+
+@pytest.mark.asyncio
+async def test_messagelist_integrity(tmp_path):
+    from drydock.core.types import MessageList
+    agent = _agent(tmp_path)
+    await _run(agent, "Hello")
+    assert isinstance(agent.messages, MessageList)
+
+
+# ============================================================================
+# Agent behavior
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_ambiguous_prompt_no_infinite_loop(tmp_path):
+    agent = _agent(tmp_path, max_turns=8)
+    events = await _run(agent, "test", max_events=50)
+    tool_calls = [e for e in events if isinstance(e, ToolCallEvent)]
+    assert len(tool_calls) < 20, f"Ambiguous 'test' caused {len(tool_calls)} tool calls"
+
+@pytest.mark.asyncio
+async def test_agent_stops_cleanly(tmp_path):
+    agent = _agent(tmp_path)
+    events = await _run(agent, "What is 2+2?", max_events=20)
+    assert len(events) < 20
+
+@pytest.mark.asyncio
+async def test_system_prompt_present(tmp_path):
+    agent = _agent(tmp_path)
+    await _run(agent, "hi")
+    assert agent.messages[0].role == Role.system
+    assert "DryDock" in agent.messages[0].content
+
+@pytest.mark.asyncio
+async def test_no_understood_in_messages(tmp_path):
+    agent = _agent(tmp_path)
+    events = await _run(agent, "Read /nonexistent/file.py")
+    for msg in agent.messages:
+        if msg.role == Role.assistant:
+            assert "Understood." not in (msg.content or "")
+
+
+# ============================================================================
+# Bash abuse prevention
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_bash_abuse_limited(tmp_path):
+    """Agent should not run 12+ bash commands without making an edit."""
+    agent = _agent(tmp_path, max_turns=15)
+    events = await _run(agent, "Explore the entire filesystem structure of this project", max_events=80)
+    bash_calls = [e for e in events if isinstance(e, ToolCallEvent) and e.tool_name == "bash"]
+    # Should be stopped before 12
+    assert len(bash_calls) <= 15, f"Agent ran {len(bash_calls)} bash commands"
+
+
+# ============================================================================
+# New tools work
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_glob_tool_works(tmp_path):
+    """Glob should find files."""
+    (tmp_path / "test1.py").write_text("# test")
+    (tmp_path / "test2.py").write_text("# test")
+    agent = _agent(tmp_path)
+    events = await _run(agent, f"Find all .py files in {tmp_path}")
+    assert len(events) >= 1

drydock_cli-1.1.2/tests/test_smoke.py

@@ -0,0 +1,173 @@
+"""SMOKE TESTS — Run on every code change (< 5 seconds).
+
+Absolute minimum checks that DryDock isn't broken.
+No backend needed. Pure code/config validation.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+
+# ============================================================================
+# 1. Package imports don't crash
+# ============================================================================
+
+class TestImports:
+    def test_core_imports(self):
+        from drydock.core.agent_loop import AgentLoop
+        from drydock.core.types import LLMMessage, MessageList, Role
+        assert AgentLoop
+        assert LLMMessage
+        assert MessageList
+
+    def test_config_imports(self):
+        from drydock.core.config import VibeConfig, ModelConfig, ProviderConfig
+        assert VibeConfig
+
+    def test_tool_imports(self):
+        from drydock.core.tools.builtins.bash import Bash
+        from drydock.core.tools.builtins.grep import Grep
+        from drydock.core.tools.builtins.search_replace import SearchReplace
+        from drydock.core.tools.builtins.glob_tool import Glob
+        from drydock.core.tools.builtins.notebook_edit import NotebookEdit
+        from drydock.core.tools.builtins.task_manager import TaskCreate
+        from drydock.core.tools.builtins.lsp import Lsp
+        from drydock.core.tools.builtins.cron import CronCreate
+        from drydock.core.tools.builtins.tool_search import ToolSearch
+        from drydock.core.tools.builtins.worktree import EnterWorktree
+        from drydock.core.tools.builtins.invoke_skill import InvokeSkill
+        assert Glob and Lsp and ToolSearch
+
+    def test_hook_imports(self):
+        from drydock.core.hooks import HookManager, HookEvent
+        assert HookManager
+        assert HookEvent.PRE_TOOL_USE
+
+    def test_consultant_imports(self):
+        from drydock.core.consultant import ask_consultant, is_consultant_available
+        assert ask_consultant
+        assert is_consultant_available
+
+
+# ============================================================================
+# 2. Branding correct
+# ============================================================================
+
+class TestBranding:
+    def test_system_prompt_says_drydock(self):
+        prompt = Path(__file__).parent.parent / "drydock" / "core" / "prompts" / "cli.md"
+        assert "DryDock" in prompt.read_text()
+
+    def test_no_vibe_in_defaults(self):
+        from drydock.core.config._settings import DEFAULT_MODELS
+        for m in DEFAULT_MODELS:
+            assert "vibe" not in m.name.lower()
+
+    def test_version_not_empty(self):
+        from drydock import __version__
+        assert __version__ and len(__version__) > 2
+
+
+# ============================================================================
+# 3. Critical safety checks
+# ============================================================================
+
+class TestSafety:
+    def test_circuit_breaker_exists(self):
+        from drydock.core.agent_loop import AgentLoop
+        assert hasattr(AgentLoop, "_circuit_breaker_check")
+
+    def test_message_ordering_exists(self):
+        from drydock.core.agent_loop import AgentLoop
+        assert hasattr(AgentLoop, "_sanitize_message_ordering")
+
+    def test_injection_guard_exists(self):
+        from drydock.core.tools.injection_guard import check_content_for_injection
+        assert check_content_for_injection("normal text") is None
+
+    def test_binary_file_guard(self):
+        from drydock.core.tools.builtins.write_file import WriteFile
+        assert hasattr(WriteFile, "_BINARY_EXTENSIONS")
+        assert ".pptx" in WriteFile._BINARY_EXTENSIONS
+
+    def test_test_file_blocked(self):
+        from drydock.core.tools.base import ToolError
+        from drydock.core.tools.builtins.search_replace import SearchReplace
+        import inspect
+        src = inspect.getsource(SearchReplace.run)
+        assert "BLOCKED" in src  # Test files should be blocked
+
+
+# ============================================================================
+# 4. Tools have prompts
+# ============================================================================
+
+class TestToolPrompts:
+    def test_all_tools_have_prompts(self):
+        tools_dir = Path(__file__).parent.parent / "drydock" / "core" / "tools" / "builtins"
+        prompts_dir = tools_dir / "prompts"
+        tool_files = {f.stem for f in tools_dir.glob("*.py") if f.stem != "__init__"}
+        prompt_files = {f.stem for f in prompts_dir.glob("*.md") if f.stem != "__init__"}
+
+        # Not all tools need prompts (some are internal)
+        core_tools = {"bash", "grep", "read_file", "search_replace", "write_file",
+                      "glob", "notebook_edit", "task_manager", "lsp", "cron",
+                      "invoke_skill", "worktree", "tool_search"}
+        missing = core_tools - prompt_files
+        assert not missing, f"Missing prompt files for: {missing}"
+
+
+# ============================================================================
+# 5. Skills bundled
+# ============================================================================
+
+class TestSkills:
+    def test_all_skills_exist(self):
+        skills_dir = Path(__file__).parent.parent / "drydock" / "skills"
+        expected = {"create-presentation", "deep-research", "investigate", "review", "ship", "batch", "simplify"}
+        actual = {d.name for d in skills_dir.iterdir() if d.is_dir() and (d / "SKILL.md").exists()}
+        missing = expected - actual
+        assert not missing, f"Missing skills: {missing}"
+
+    def test_skills_have_valid_frontmatter(self):
+        from drydock.core.skills.parser import parse_frontmatter
+        skills_dir = Path(__file__).parent.parent / "drydock" / "skills"
+        for skill_dir in skills_dir.iterdir():
+            skill_file = skill_dir / "SKILL.md"
+            if skill_file.exists():
+                content = skill_file.read_text()
+                metadata, _ = parse_frontmatter(content)
+                assert "name" in metadata, f"Skill {skill_dir.name} missing 'name'"
+
+
+# ============================================================================
+# 6. Config
+# ============================================================================
+
+class TestConfig:
+    def test_consultant_model_field(self):
+        from drydock.core.config import VibeConfig
+        assert "consultant_model" in VibeConfig.model_fields
+
+    def test_drydock_home_default(self):
+        import inspect
+        from drydock.core.paths import _vibe_home
+        src = inspect.getsource(_vibe_home)
+        assert ".drydock" in src
+
+
+# ============================================================================
+# 7. Commands registered
+# ============================================================================
+
+class TestCommands:
+    def test_core_commands_registered(self):
+        from drydock.cli.commands import CommandRegistry
+        registry = CommandRegistry()
+        expected = {"help", "config", "clear", "compact", "exit", "status", "consult", "rewind"}
+        actual = set(registry.commands.keys())
+        missing = expected - actual
+        assert not missing, f"Missing commands: {missing}"