drydock-cli 0.5.2__tar.gz → 0.6.1__tar.gz

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 0.5.2
+Version: 0.6.1
 Summary: Drydock — nautical CLI coding agent
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/PRD.md

@@ -153,10 +153,32 @@ User-reported issues from real usage testing, all fixed:
 
 ### Phase 5b: Testing & CI (Mar 24)
 
-- **64 regression tests** in 2 test files (0.12s total runtime)
-- **Tests gate all deploys** — deploy_to_github.sh and publish_to_pypi.sh abort on failure
-- **Continuous SWE-bench loop** — 320/500 tasks tested, 45.3% pass rate, auto-restarts on reboot
-- **PyPI published** — drydock-cli v0.4.1
+- **76 regression tests** gating all deploys
+- **Continuous SWE-bench loop** — 500/500 tasks tested, 49.0% pass rate (+9.4% from baseline)
+- **vibe/ → drydock/** package rename (908 imports updated)
+- **PyPI published** — drydock-cli v0.5.2
+
+### Phase 6: GSD Integration & Performance (Mar 25)
+
+GSD-inspired features (from [get-shit-done](https://github.com/gsd-build/get-shit-done)):
+
+| Feature | Details |
+|---------|---------|
+| Tiered context warnings | 4 levels at 50/65/75/85% usage, debounced every 5 calls |
+| Prompt injection guard | Detects role overrides, invisible Unicode, hidden instructions in file writes |
+| Structured state file | `.drydock/state.md` persists task, files, decisions across sessions |
+| Deviation handling rules | Auto-fix bugs/imports, ask user for architecture/scope decisions |
+
+Performance and loop fixes:
+
+| Fix | Details |
+|-----|---------|
+| Circuit breaker | Blocks exact same tool call after 2 attempts, returns cached result |
+| Thinking flicker fix | Status words change every 4s, not every token |
+| Conda env protection | Preserves user's active conda environment in subprocesses |
+| `--insecure` / `-k` flag | Disables SSL verification for corporate proxy environments |
+| `--consultant MODEL` flag | Allows calling a smarter model for single-turn advice when stuck |
+| 81 regression tests | Circuit breaker, injection guard, state file, context tiers all tested |
 
 ---
 

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/drydock/cli/entrypoint.py

@@ -90,6 +90,19 @@ def parse_arguments() -> argparse.Namespace:
         help="Skip all tool permission checks. Equivalent to --agent auto-approve. "
         "Use with caution — tools will execute without confirmation.",
     )
+    parser.add_argument(
+        "-k", "--insecure",
+        action="store_true",
+        help="Disable SSL certificate verification for web searches and API calls. "
+        "Useful behind corporate proxies with self-signed certificates.",
+    )
+    parser.add_argument(
+        "--consultant",
+        metavar="MODEL",
+        help="Enable consultant mode: when the agent is uncertain or stuck in a loop, "
+        "it can call a more capable model (e.g., 'gemini-2.5-pro') for single-turn advice. "
+        "The consultant is NOT used for tool calls, only reasoning.",
+    )
 
     # Feature flag for teleport, not exposed to the user yet
     parser.add_argument("--teleport", action="store_true", help=argparse.SUPPRESS)
@@ -160,6 +173,18 @@ def main() -> None:
     if args.dangerously_skip_permissions:
         args.agent = BuiltinAgentName.AUTO_APPROVE
 
+    # --insecure → disable SSL verification globally
+    if args.insecure:
+        os.environ["DRYDOCK_INSECURE"] = "1"
+        os.environ["CURL_CA_BUNDLE"] = ""
+        os.environ["REQUESTS_CA_BUNDLE"] = ""
+        import ssl
+        ssl._create_default_https_context = ssl._create_unverified_context
+
+    # --consultant → store for agent loop to use
+    if args.consultant:
+        os.environ["DRYDOCK_CONSULTANT_MODEL"] = args.consultant
+
     is_interactive = args.prompt is None
     if is_interactive:
         check_and_resolve_trusted_folder()

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/drydock/cli/textual_ui/widgets/loading.py

@@ -64,6 +64,9 @@ class LoadingWidget(SpinnerMixin, Static):
         "Stocking the galley",
     ]
 
+    # Minimum seconds between status word changes (prevents flicker)
+    _STATUS_CHANGE_INTERVAL = 4.0
+
     def __init__(self, status: str | None = None) -> None:
         super().__init__(classes="loading-widget")
         self.init_spinner()
@@ -77,6 +80,7 @@ class LoadingWidget(SpinnerMixin, Static):
         self._last_elapsed: int = -1
         self._paused_total: float = 0.0
         self._pause_start: float | None = None
+        self._last_status_change: float = 0.0
 
     def _get_easter_egg(self) -> str | None:
         EASTER_EGG_PROBABILITY = 0.10
@@ -111,6 +115,11 @@ class LoadingWidget(SpinnerMixin, Static):
             self._pause_start = None
 
     def set_status(self, status: str) -> None:
+        # Throttle status word changes to prevent flicker
+        now = time()
+        if now - self._last_status_change < self._STATUS_CHANGE_INTERVAL:
+            return  # Too soon — keep current word
+        self._last_status_change = now
         self.status = self._apply_easter_egg(status)
         self._update_animation()
 

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/drydock/core/agent_loop.py

@@ -188,6 +188,10 @@ class AgentLoop:
         self.middleware_pipeline = MiddlewarePipeline()
         self._setup_middleware()
 
+        # Circuit breaker: track tool call signatures to prevent exact repeats
+        # Key: hash(tool_name + args), Value: (count, last_result_snippet)
+        self._tool_call_history: dict[str, tuple[int, str]] = {}
+
         system_prompt = get_universal_system_prompt(
             self.tool_manager, self.config, self.skill_manager, self.agent_manager
         )
@@ -831,9 +835,81 @@ class AgentLoop:
                 )
             )
 
+    def _get_attempted_summary(self) -> str:
+        """Build a summary of what the agent has already tried."""
+        if not self._tool_call_history:
+            return ""
+        attempts = []
+        for sig, (count, result) in self._tool_call_history.items():
+            if count >= 2:
+                attempts.append(f"  - Ran {count}x, result: {result[:80]}")
+        if attempts:
+            return "ALREADY ATTEMPTED (do NOT repeat):\n" + "\n".join(attempts[:10])
+        return ""
+
+    async def _circuit_breaker_check(self, tool_call: ResolvedToolCall) -> str | None:
+        """Block exact-duplicate tool calls. Returns cached result or None.
+        On 3rd attempt, auto-consults the smarter model if available."""
+        sig = hashlib.md5(
+            f"{tool_call.tool_name}:{tool_call.raw_arguments}".encode()
+        ).hexdigest()
+
+        count, last_result = self._tool_call_history.get(sig, (0, ""))
+        if count >= 2:
+            attempted = self._get_attempted_summary()
+            msg = (
+                f"CIRCUIT BREAKER: You already ran `{tool_call.tool_name}` with these "
+                f"exact arguments {count} times and got the same result each time.\n\n"
+                f"Previous result: {last_result[:200]}\n\n"
+                f"{attempted}\n\n"
+                f"STOP repeating. You MUST try something DIFFERENT:\n"
+                f"- Different arguments or search terms\n"
+                f"- A completely different tool\n"
+                f"- Ask the user for clarification"
+            )
+
+            # Auto-consult smarter model if available
+            if count == 2:
+                try:
+                    from drydock.core.consultant import is_consultant_available, ask_consultant
+                    if is_consultant_available():
+                        question = (
+                            f"I'm stuck in a loop. I keep running `{tool_call.tool_name}` "
+                            f"with args: {tool_call.raw_arguments[:100]} "
+                            f"and getting: {last_result[:200]}. "
+                            f"What should I do differently?"
+                        )
+                        advice = await ask_consultant(question)
+                        if advice:
+                            msg += f"\n\nCONSULTANT ADVICE: {advice}"
+                except Exception as e:
+                    logger.debug("Consultant call failed: %s", e)
+
+            return msg
+        return None
+
+    def _circuit_breaker_record(self, tool_call: ResolvedToolCall, result_text: str) -> None:
+        """Record a tool call execution for circuit breaker tracking."""
+        sig = hashlib.md5(
+            f"{tool_call.tool_name}:{tool_call.raw_arguments}".encode()
+        ).hexdigest()
+        count, _ = self._tool_call_history.get(sig, (0, ""))
+        self._tool_call_history[sig] = (count + 1, result_text[:500])
+
     async def _process_one_tool_call(
         self, tool_call: ResolvedToolCall
     ) -> AsyncGenerator[ToolResultEvent | ToolStreamEvent]:
+        # Circuit breaker: block exact duplicate calls after 2 attempts
+        if blocked := await self._circuit_breaker_check(tool_call):
+            yield ToolResultEvent(
+                tool_name=tool_call.tool_name,
+                tool_class=tool_call.tool_class,
+                error=blocked,
+                tool_call_id=tool_call.call_id,
+            )
+            self._handle_tool_response(tool_call, blocked, "failure")
+            return
+
         try:
             tool_instance = self.tool_manager.get(tool_call.tool_name)
         except Exception as exc:
@@ -902,6 +978,8 @@ class AgentLoop:
 
             result_dict = result_model.model_dump()
             text = "\n".join(f"{k}: {v}" for k, v in result_dict.items())
+            # Record for circuit breaker
+            self._circuit_breaker_record(tool_call, text)
             self._handle_tool_response(
                 tool_call, text, "success", decision, result_dict
             )

drydock_cli-0.6.1/drydock/core/consultant.py

@@ -0,0 +1,141 @@
+"""Consultant agent — asks a smarter model for single-turn advice.
+
+The consultant is a READ-ONLY advisor:
+- The local model stays in control of ALL tool calls
+- The consultant NEVER calls tools, writes files, or runs commands
+- It only receives a question and returns reasoning/advice
+- Used when the local model is stuck, uncertain, or looping
+
+Usage from agent_loop:
+    advice = await ask_consultant("I found 3 files that match. Which one likely contains the bug?")
+
+The consultant model is configured via:
+- CLI: drydock --consultant gemini-2.5-pro
+- Env: DRYDOCK_CONSULTANT_MODEL=gemini-2.5-pro
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+from typing import Any
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+
+async def ask_consultant(
+    question: str,
+    context: str = "",
+    model: str | None = None,
+    api_key: str | None = None,
+    api_base: str | None = None,
+) -> str:
+    """Ask the consultant model a question and get text advice back.
+
+    Parameters
+    ----------
+    question : str
+        The specific question to ask.
+    context : str
+        Optional context (e.g., file contents, error messages).
+    model : str
+        Model name. Defaults to DRYDOCK_CONSULTANT_MODEL env var.
+    api_key : str
+        API key. Auto-detected from common env vars if not provided.
+    api_base : str
+        API base URL. Auto-detected based on model name if not provided.
+
+    Returns
+    -------
+    str
+        The consultant's advice text, or an error message.
+    """
+    model = model or os.environ.get("DRYDOCK_CONSULTANT_MODEL", "")
+    if not model:
+        return ""
+
+    # Auto-detect API key and base URL from model name
+    api_key = api_key or _detect_api_key(model)
+    api_base = api_base or _detect_api_base(model)
+
+    if not api_key:
+        logger.warning("No API key found for consultant model '%s'", model)
+        return ""
+
+    prompt = f"""You are a senior software engineering consultant. A junior AI coding agent is asking for your advice.
+Give concise, actionable guidance. Do NOT provide code — just tell them WHAT to do and WHERE to look.
+
+QUESTION:
+{question}
+"""
+    if context:
+        prompt += f"\nCONTEXT:\n{context[:2000]}\n"
+
+    try:
+        # Use OpenAI-compatible API (works with Gemini, OpenAI, Mistral, local models)
+        verify_ssl = os.environ.get("DRYDOCK_INSECURE") != "1"
+        async with httpx.AsyncClient(verify=verify_ssl, timeout=30.0) as client:
+            response = await client.post(
+                f"{api_base}/chat/completions",
+                headers={
+                    "Authorization": f"Bearer {api_key}",
+                    "Content-Type": "application/json",
+                },
+                json={
+                    "model": model,
+                    "messages": [{"role": "user", "content": prompt}],
+                    "max_tokens": 500,
+                    "temperature": 0.3,
+                    # NO tools — consultant is read-only
+                },
+            )
+            response.raise_for_status()
+            data = response.json()
+            advice = data["choices"][0]["message"]["content"]
+            logger.info("Consultant (%s) responded: %s", model, advice[:100])
+            return advice
+
+    except Exception as e:
+        logger.warning("Consultant call failed: %s", e)
+        return f"(Consultant unavailable: {e})"
+
+
+def _detect_api_key(model: str) -> str:
+    """Auto-detect API key from environment based on model name."""
+    model_lower = model.lower()
+
+    if "gemini" in model_lower:
+        return os.environ.get("GEMINI_API_KEY", os.environ.get("GOOGLE_API_KEY", ""))
+    if "gpt" in model_lower or "o1" in model_lower or "o3" in model_lower:
+        return os.environ.get("OPENAI_API_KEY", "")
+    if "claude" in model_lower:
+        return os.environ.get("ANTHROPIC_API_KEY", "")
+    if "mistral" in model_lower or "devstral" in model_lower:
+        return os.environ.get("MISTRAL_API_KEY", "")
+
+    # Generic fallback
+    return os.environ.get("DRYDOCK_CONSULTANT_API_KEY", "")
+
+
+def _detect_api_base(model: str) -> str:
+    """Auto-detect API base URL from model name."""
+    model_lower = model.lower()
+
+    if "gemini" in model_lower:
+        return "https://generativelanguage.googleapis.com/v1beta/openai"
+    if "gpt" in model_lower or "o1" in model_lower or "o3" in model_lower:
+        return "https://api.openai.com/v1"
+    if "claude" in model_lower:
+        return "https://api.anthropic.com/v1"
+    if "mistral" in model_lower or "devstral" in model_lower:
+        return os.environ.get("MISTRAL_API_BASE", "https://api.mistral.ai/v1")
+
+    # Try local vLLM
+    return os.environ.get("DRYDOCK_CONSULTANT_API_BASE", "http://localhost:8000/v1")
+
+
+def is_consultant_available() -> bool:
+    """Check if a consultant model is configured."""
+    return bool(os.environ.get("DRYDOCK_CONSULTANT_MODEL"))

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/drydock/core/tools/builtins/bash.py

@@ -110,11 +110,25 @@ def _get_base_env() -> dict[str, str]:
 
         # Enable conda in non-interactive shells by setting BASH_ENV
         # to source conda's setup script. Without this, `conda activate`
-        # fails because it's a shell function defined in .bashrc (which
-        # isn't sourced for non-interactive shell commands).
+        # fails because it's a shell function defined in .bashrc.
+        #
+        # IMPORTANT: We also preserve the user's active conda env to avoid
+        # interfering with other environments and losing their aliases.
         conda_sh = _get_conda_setup_script()
         if conda_sh:
-            base_env["BASH_ENV"] = conda_sh
+            # Create a wrapper that sources conda.sh then re-activates the user's env
+            active_env = os.environ.get("CONDA_DEFAULT_ENV", "")
+            if active_env and active_env != "base":
+                # Write a temp script that initializes conda AND activates the user's env
+                import tempfile
+                wrapper = tempfile.NamedTemporaryFile(
+                    mode="w", suffix=".sh", delete=False, prefix="drydock_conda_"
+                )
+                wrapper.write(f'source "{conda_sh}"\nconda activate "{active_env}" 2>/dev/null\n')
+                wrapper.close()
+                base_env["BASH_ENV"] = wrapper.name
+            else:
+                base_env["BASH_ENV"] = conda_sh
 
     return base_env
 

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "0.5.2"
+version = "0.6.1"
 description = "Drydock — nautical CLI coding agent"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-0.5.2 → drydock_cli-0.6.1}/tests/test_drydock_tasks.py

@@ -16,6 +16,7 @@ These catch the issues users find when actually using the application:
 from __future__ import annotations
 
 from pathlib import Path
+from unittest.mock import patch
 
 import pytest
 
@@ -382,6 +383,84 @@ class TestDeviationHandling:
         assert "Auto-fix" in content or "Auto-resolve" in content
 
 
+# ============================================================================
+# Circuit Breaker
+# ============================================================================
+
+class TestCircuitBreaker:
+    """Prevents exact duplicate tool calls."""
+
+    def test_circuit_breaker_blocks_after_2(self):
+        import asyncio
+        from drydock.core.agent_loop import AgentLoop
+        from drydock.core.types import MessageList
+        from types import SimpleNamespace
+
+        al = object.__new__(AgentLoop)
+        al.messages = MessageList()
+        al._tool_call_history = {}
+
+        tc = SimpleNamespace(tool_name="bash", raw_arguments='{"command": "ls -ltr"}')
+
+        assert asyncio.run(al._circuit_breaker_check(tc)) is None
+        al._circuit_breaker_record(tc, "file1.py\nfile2.py")
+        assert asyncio.run(al._circuit_breaker_check(tc)) is None
+        al._circuit_breaker_record(tc, "file1.py\nfile2.py")
+
+        result = asyncio.run(al._circuit_breaker_check(tc))
+        assert result is not None
+        assert "CIRCUIT BREAKER" in result
+
+    def test_different_args_not_blocked(self):
+        import asyncio
+        from drydock.core.agent_loop import AgentLoop
+        from drydock.core.types import MessageList
+        from types import SimpleNamespace
+
+        al = object.__new__(AgentLoop)
+        al.messages = MessageList()
+        al._tool_call_history = {}
+
+        tc1 = SimpleNamespace(tool_name="bash", raw_arguments='{"command": "ls -ltr"}')
+        tc2 = SimpleNamespace(tool_name="bash", raw_arguments='{"command": "ls -la"}')
+
+        al._circuit_breaker_record(tc1, "result1")
+        al._circuit_breaker_record(tc1, "result1")
+        assert asyncio.run(al._circuit_breaker_check(tc1)) is not None
+        assert asyncio.run(al._circuit_breaker_check(tc2)) is None
+
+
+# ============================================================================
+# Insecure Flag
+# ============================================================================
+
+class TestInsecureFlag:
+    def test_insecure_flag_exists(self):
+        from drydock.cli.entrypoint import parse_arguments
+        import sys
+        with patch.object(sys, "argv", ["drydock", "-k", "-p", "test"]):
+            args = parse_arguments()
+            assert args.insecure is True
+
+    def test_consultant_flag_exists(self):
+        from drydock.cli.entrypoint import parse_arguments
+        import sys
+        with patch.object(sys, "argv", ["drydock", "--consultant", "gemini-2.5-pro", "-p", "test"]):
+            args = parse_arguments()
+            assert args.consultant == "gemini-2.5-pro"
+
+
+# ============================================================================
+# Thinking Flicker Throttle
+# ============================================================================
+
+class TestThinkingThrottle:
+    def test_status_change_interval_exists(self):
+        from drydock.cli.textual_ui.widgets.loading import LoadingWidget
+        assert hasattr(LoadingWidget, "_STATUS_CHANGE_INTERVAL")
+        assert LoadingWidget._STATUS_CHANGE_INTERVAL >= 3.0
+
+
 class TestEasterEggs:
     """Easter eggs are nautical, not French."""