drydock-cli 0.6.7__tar.gz → 0.7.1__tar.gz

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 0.6.7
+Version: 0.7.1
 Summary: Drydock — nautical CLI coding agent
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/drydock/__init__.py

@@ -4,4 +4,4 @@ from pathlib import Path
 
 DRYDOCK_ROOT = Path(__file__).parent
 VIBE_ROOT = DRYDOCK_ROOT  # Backward compatibility alias
-__version__ = "0.5.0"
+__version__ = "0.7.0"

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/drydock/core/prompts/cli.md

@@ -174,13 +174,15 @@ Two-Phase Workflow (mandatory):
 PHASE 1 — INVESTIGATE (2-3 tool calls, no edits):
 Use the failing test path to identify the module (e.g. `tests/models/test_query.py` → search in `models/`).
 grep for the specific class/function from the bug report, SCOPED to that module.
-read_file the ONE most relevant source function (50-100 lines with offset/limit).
+read_file the most relevant source function(s) (50-100 lines with offset/limit).
 State your findings:
-  TARGET: path/to/file.py
+  TARGET: path/to/file.py (and path/to/other.py if the fix spans multiple files)
   FUNCTION: function_name
   CAUSE: one sentence root cause
   FIX: one sentence fix approach
 
+If the bug involves multiple connected components (e.g., a model field AND a query compiler, or a config parser AND a validator), you MAY edit multiple source files. But keep each edit minimal.
+
 PHASE 2 — FIX (after investigation):
 Go directly to the TARGET file. search_replace to fix. Read back changed lines to verify.
 

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/drydock/core/tools/builtins/prompts/swe_bench.md

@@ -9,11 +9,12 @@ WORKFLOW:
 
 CRITICAL RULES:
 - When grep returns results in both source and test files, ALWAYS look at source files first
-- Most bugs need changes to ONE file only — don't over-engineer
+- Most bugs need 1 file change, but some need 2-3 related files — fix all of them
 - Keep search_replace old_str to 1-5 lines — just the lines being changed plus unique context
 - If search_replace fails with "not found", re-read the file to get the EXACT current text
 - Do NOT run tests — the test runner handles that separately
 - After your edit, verify by reading back the modified lines
+- If the bug involves TWO related components (e.g., serializer + deserializer, model + migration), check if BOTH need changes
 
 FILE DISCOVERY:
 - Extract the module path from the test path: `tests/models/test_query.py` → search in `models/`

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "0.6.7"
+version = "0.7.1"
 description = "Drydock — nautical CLI coding agent"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/scripts/deploy_to_github.sh

@@ -37,7 +37,7 @@ AUTH_URL="https://${GITHUB_TOKEN}@github.com/fbobe321/drydock.git"
 
 log "Running regression tests..."
 cd "$DRYDOCK_SRC"
-if ! python3 -m pytest tests/test_drydock_regression.py tests/test_drydock_tasks.py tests/test_loop_detection.py tests/test_agent_tasks.py \
+if ! python3 -m pytest tests/test_drydock_regression.py tests/test_drydock_tasks.py tests/test_loop_detection.py tests/test_agent_tasks.py tests/test_integration.py \
     -p no:xdist -p no:cov --override-ini="addopts=" -q 2>&1; then
     log "TESTS FAILED — deploy aborted. Fix the tests before deploying."
     exit 1

{drydock_cli-0.6.7 → drydock_cli-0.7.1}/scripts/publish_to_pypi.sh

@@ -25,7 +25,7 @@ PYPI_TOKEN=$(cat "$TOKEN_FILE")
 
 # --- Run tests ---
 log "Running regression tests..."
-if ! $PYTHON -m pytest tests/test_drydock_regression.py tests/test_drydock_tasks.py tests/test_loop_detection.py tests/test_agent_tasks.py \
+if ! $PYTHON -m pytest tests/test_drydock_regression.py tests/test_drydock_tasks.py tests/test_loop_detection.py tests/test_agent_tasks.py tests/test_integration.py \
     -p no:xdist -p no:cov --override-ini="addopts=" -q 2>&1; then
     log "TESTS FAILED — publish aborted."
     exit 1

drydock_cli-0.7.1/tests/test_integration.py

@@ -0,0 +1,270 @@
+"""Integration tests — use the REAL vLLM backend.
+
+These tests hit the actual LLM at localhost:8000 and verify the full
+pipeline works end-to-end. They catch issues that mock tests miss
+(like the raw_arguments crash that broke 159 tasks).
+
+Skip if vLLM is not running (CI environments).
+
+Run manually: pytest tests/test_integration.py -v -s
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import os
+from pathlib import Path
+
+import httpx
+import pytest
+
+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,
+    ToolCallEvent,
+    ToolResultEvent,
+    UserMessageEvent,
+)
+
+
+# ============================================================================
+# Skip if vLLM not running
+# ============================================================================
+
+def _vllm_available() -> bool:
+    try:
+        r = httpx.get("http://localhost:8000/v1/models", timeout=3)
+        return r.status_code == 200
+    except Exception:
+        return False
+
+
+pytestmark = pytest.mark.skipif(
+    not _vllm_available(),
+    reason="vLLM not running at localhost:8000"
+)
+
+
+# ============================================================================
+# Helpers
+# ============================================================================
+
+def _make_config(tmp_path: Path) -> VibeConfig:
+    """Create a config that uses local vLLM."""
+    return VibeConfig(
+        active_model="devstral",
+        auto_approve=True,
+        enable_telemetry=False,
+        include_project_context=False,
+        include_prompt_detail=False,
+        system_prompt_id="tests",
+        providers=[
+            ProviderConfig(
+                name="local-vllm",
+                api_base="http://localhost:8000/v1",
+                api_key_env_var="",
+                backend=Backend.GENERIC,
+            ),
+        ],
+        models=[
+            ModelConfig(
+                name="devstral",
+                provider="local-vllm",
+                alias="devstral",
+                input_price=0.0,
+                output_price=0.0,
+            ),
+        ],
+        session_logging={"enabled": False, "save_dir": str(tmp_path / "logs")},
+    )
+
+
+def _make_agent(tmp_path: Path) -> AgentLoop:
+    config = _make_config(tmp_path)
+    return AgentLoop(
+        config=config,
+        agent_name=BuiltinAgentName.AUTO_APPROVE,
+        max_turns=5,  # Safety limit
+    )
+
+
+async def _run(agent: AgentLoop, prompt: str, max_events: int = 100) -> list[BaseEvent]:
+    events = []
+    async for ev in agent.act(prompt):
+        events.append(ev)
+        if len(events) >= max_events:
+            break
+    return events
+
+
+# ============================================================================
+# Test 1: Simple text response from real LLM
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_real_llm_responds(tmp_path):
+    """Real LLM produces a text response without crashing."""
+    agent = _make_agent(tmp_path)
+    events = await _run(agent, "Say hello in one word.")
+
+    assert any(isinstance(e, AssistantEvent) for e in events)
+    text = [e for e in events if isinstance(e, AssistantEvent) and e.content]
+    assert len(text) >= 1
+
+
+# ============================================================================
+# Test 2: LLM calls a tool and gets result
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_real_tool_call(tmp_path):
+    """Real LLM calls a tool (e.g., bash/grep) and receives the result."""
+    agent = _make_agent(tmp_path)
+    events = await _run(agent, "Run: echo 'drydock test passed'")
+
+    # Should have at least one tool call or text response
+    has_tool = any(isinstance(e, ToolCallEvent) for e in events)
+    has_text = any(isinstance(e, AssistantEvent) and e.content for e in events)
+    assert has_tool or has_text, "LLM produced neither tool calls nor text"
+
+
+# ============================================================================
+# Test 3: No crash on tool execution path
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_no_crash_on_tool_path(tmp_path):
+    """The full tool execution path (resolve → execute → result) doesn't crash."""
+    agent = _make_agent(tmp_path)
+
+    # This prompt should trigger grep or read_file
+    events = await _run(agent, "What files are in the current directory? Use ls.")
+
+    # If we get any events at all without exception, the path works
+    assert len(events) >= 1
+
+    # No tool results should have crash errors
+    for e in events:
+        if isinstance(e, ToolResultEvent) and e.error:
+            assert "AttributeError" not in e.error, f"Crash in tool path: {e.error}"
+            assert "raw_arguments" not in e.error, f"Old bug present: {e.error}"
+
+
+# ============================================================================
+# Test 4: Circuit breaker doesn't crash on real tool calls
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_circuit_breaker_no_crash(tmp_path):
+    """Circuit breaker works with real ResolvedToolCall objects."""
+    agent = _make_agent(tmp_path)
+
+    # Ask something that triggers a tool call
+    events = await _run(agent, "Read the first 3 lines of /etc/hostname")
+
+    # Verify circuit breaker state was populated (even if only 1 call)
+    # The key thing is it didn't crash
+    assert hasattr(agent, "_tool_call_history")
+    assert isinstance(agent._tool_call_history, dict)
+
+
+# ============================================================================
+# Test 5: Message ordering stays clean after real conversation
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_message_ordering_real(tmp_path):
+    """No user-after-tool violations after a real multi-turn conversation."""
+    agent = _make_agent(tmp_path)
+    events = await _run(agent, "What is 2+2? Just answer the number.")
+
+    from drydock.core.types import Role
+    for i in range(1, len(agent.messages)):
+        if agent.messages[i].role == Role.user and agent.messages[i - 1].role == Role.tool:
+            pytest.fail(f"user after tool at message {i}")
+
+
+# ============================================================================
+# Test 6: Agent stops within turn limit
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_turn_limit_respected(tmp_path):
+    """Agent respects max_turns and doesn't loop forever."""
+    agent = _make_agent(tmp_path)  # max_turns=5
+    events = await _run(agent, "Explore this entire codebase thoroughly.", max_events=200)
+
+    # Should stop eventually (max_turns=5 or natural stop)
+    assert len(events) < 200, "Agent seems to be looping"
+
+
+# ============================================================================
+# Test 7: Bash tool works end-to-end
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_bash_tool_works(tmp_path):
+    """Bash tool executes and returns output through the full pipeline."""
+    agent = _make_agent(tmp_path)
+    events = await _run(agent, "Run this exact command: echo 'hello from drydock'")
+
+    tool_results = [e for e in events if isinstance(e, ToolResultEvent)]
+    # Either the model called bash successfully, or gave a text response
+    has_any_output = len(tool_results) > 0 or any(
+        isinstance(e, AssistantEvent) and e.content for e in events
+    )
+    assert has_any_output
+
+
+# ============================================================================
+# Test 8: Agent handles ambiguous prompt without infinite loop
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_ambiguous_prompt_no_loop(tmp_path):
+    """Short ambiguous prompt ('test') doesn't cause infinite filesystem exploration."""
+    agent = _make_agent(tmp_path)
+    events = await _run(agent, "test", max_events=50)
+
+    # Should not use excessive tool calls
+    tool_calls = [e for e in events if isinstance(e, ToolCallEvent)]
+    assert len(tool_calls) < 20, f"Too many tool calls ({len(tool_calls)}) for ambiguous prompt"
+
+
+# ============================================================================
+# Test 9: MessageList integrity after real execution
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_messagelist_integrity_real(tmp_path):
+    """MessageList is never replaced with plain list during real execution."""
+    agent = _make_agent(tmp_path)
+    await _run(agent, "Hello")
+
+    from drydock.core.types import MessageList
+    assert isinstance(agent.messages, MessageList), \
+        f"messages is {type(agent.messages)}, not MessageList"
+
+
+# ============================================================================
+# Test 10: Sanitize message ordering runs without error
+# ============================================================================
+
+@pytest.mark.asyncio
+async def test_sanitize_runs_on_real_messages(tmp_path):
+    """_sanitize_message_ordering doesn't crash on real message history."""
+    agent = _make_agent(tmp_path)
+    await _run(agent, "What time is it?")
+
+    # Run sanitize explicitly — should not raise
+    agent._sanitize_message_ordering()
+    assert len(agent.messages) >= 2  # At least system + user