drydock-cli 2.9.97__tar.gz → 2.9.101__tar.gz

{drydock_cli-2.9.97 → drydock_cli-2.9.101}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.9.97
+Version: 2.9.101
 Summary: Local-first CLI coding agent — tested with Gemma 4 26B via llama.cpp
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-2.9.97 → drydock_cli-2.9.101}/drydock/core/llm/grammar/policy_union.py

@@ -172,6 +172,20 @@ def select_union_grammar(
     """
     if os.environ.get("DRYDOCK_GRAMMAR_DISABLE", "").strip() == "1":
         return None
+    # Union grammar is now OFF BY DEFAULT (2026-06-06, v2.9.98).
+    # Reason: a 72-tool grammar union makes the model pick whichever
+    # branch has the simplest argument schema and iterate through them
+    # mechanically — `task` (`{task: str}`), `exit_plan_mode` (`{}`),
+    # `notebook_edit`, `count`, etc. — instead of selecting the right
+    # tool for the work. Multiple visible user-facing loops 2026-06-06.
+    # Defense in depth: keep forced-single-tool grammar (policy.py)
+    # active so that when drydock explicitly forces tool_choice=write_file
+    # we still get JSON-validity guarantees. Auto-mode falls back to
+    # the model's native tool-call decisions, which Gemma 4 handles
+    # well via llama.cpp `--jinja`.
+    # Re-enable by setting DRYDOCK_UNION_GRAMMAR_ENABLE=1.
+    if os.environ.get("DRYDOCK_UNION_GRAMMAR_ENABLE", "").strip() != "1":
+        return None
     if os.environ.get("DRYDOCK_UNION_GRAMMAR_DISABLE", "").strip() == "1":
         return None
 

{drydock_cli-2.9.97 → drydock_cli-2.9.101}/drydock/core/tools/manager.py

@@ -210,6 +210,24 @@ class ToolManager:
             # inline — the auto-continue loop (tool_turns > 0 → don't
             # break) keeps it going without needing an explicit list.
             "todo",
+            # cron_* disabled for Gemma 4 (added v2.9.100). Scheduled-task
+            # management has zero relevance to coding workflows. Observed
+            # 2026-06-06 v2.9.87 batch `reshard-c4-data`: the model called
+            # cron_list 3 times in a row when stuck on a sharding problem
+            # — pure escape-hatch behavior. Operators who need cron use
+            # slash commands.
+            "cron_create", "cron_delete", "cron_list",
+            # exit_plan_mode disabled for Gemma 4 (added v2.9.98). The
+            # tool takes NO arguments (`ExitPlanModeArgs` is `pass`), so
+            # under any tool-call mode it's the absolute easiest commitment
+            # for the model. User session 2026-06-06 v2.9.97: the model
+            # called exit_plan_mode 4 times in a row, each getting
+            # "Already in implementation mode" — the no-op response is
+            # friendly enough that the model never realized it was
+            # spinning. Drydock doesn't normally use plan_mode for code
+            # work, and slash commands exist for the rare cases that
+            # need it.
+            "exit_plan_mode",
             # notebook_edit disabled for Gemma 4 (added v2.9.94 after
             # user session 2026-06-06 where the model called
             # `notebook_edit` on `/data3/slides/cli.py` — a regular .py

{drydock_cli-2.9.97 → drydock_cli-2.9.101}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.9.97"
+version = "2.9.101"
 description = "Local-first CLI coding agent — tested with Gemma 4 26B via llama.cpp"
 readme = "README.md"
 requires-python = ">=3.12"

drydock_cli-2.9.101/tests/test_gemma4_tool_filter.py

@@ -0,0 +1,88 @@
+"""Tool-list narrowing regression tests for Gemma-derived models.
+
+User-observed loops 2026-06-06:
+  - task -> blocked -> task -> blocked x 6     (v2.9.93-.97 saga)
+  - exit_plan_mode -> "Already in implementation mode" x 4
+  - notebook_edit on a .py file
+  - count tool on "."
+
+Root cause: with 72 tools exposed, the model gravitates to whichever
+tool has the trivialest arg shape. Grammar union (v2.9.86-.97) made it
+strict iteration. Even with union OFF (v2.9.98), Gemma 4 still wanders
+through too many low-value branches.
+
+These tests pin which tools must NOT be exposed for Gemma-derived
+models so the regression can't come back.
+"""
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+import pytest
+
+from drydock.core.tools.manager import ToolManager
+
+
+def _gemma4_available_names() -> set[str]:
+    """Build a ToolManager configured for a Gemma 4 model and return the
+    set of tool names the model would see in `available_tools`."""
+    cfg = MagicMock()
+    cfg.enabled_tools = None
+    cfg.disabled_tools = None
+    cfg.mcp_servers = []
+    cfg.tool_paths = []
+
+    class FakeModel:
+        name = "gemma-4-26b"
+        alias = "gemma4"
+
+    cfg.get_active_model = lambda: FakeModel()
+
+    mgr = ToolManager(config_getter=lambda: cfg)
+    return set(mgr.available_tools.keys())
+
+
+# ---------------------------------------------------------------------------
+# Tools that MUST be filtered out for Gemma 4
+# ---------------------------------------------------------------------------
+
+@pytest.mark.parametrize("forbidden", [
+    "task",            # v2.9.97: re-disabled after looping pattern
+    "task_create",
+    "task_update",
+    "task_list",
+    "ask_user_question",
+    "invoke_skill",
+    "tool_search",
+    "todo",
+    "search_replace",  # v2.9.84+: disabled due to arg-copying loops
+    "notebook_edit",   # v2.9.95: confusion bait, write_file handles .ipynb
+    "exit_plan_mode",  # v2.9.98: 0-arg "Already in implementation mode" loop
+    "cron_create",     # v2.9.100: zero relevance to coding, escape hatch
+    "cron_delete",
+    "cron_list",       # v2.9.100: 3-in-a-row in reshard-c4-data trial
+])
+def test_gemma4_does_not_expose_loop_prone_tool(forbidden):
+    names = _gemma4_available_names()
+    assert forbidden not in names, (
+        f"{forbidden!r} is exposed for Gemma 4 - known loop driver. "
+        f"Add it to _GEMMA4_AUTO_DISABLE in drydock/core/tools/manager.py."
+    )
+
+
+# ---------------------------------------------------------------------------
+# Sanity: the core write/edit set IS still exposed for Gemma 4
+# ---------------------------------------------------------------------------
+
+@pytest.mark.parametrize("required", [
+    "write_file",
+    "read_file",
+    "bash",
+    "grep",
+    "glob",
+])
+def test_gemma4_keeps_core_tools(required):
+    names = _gemma4_available_names()
+    assert required in names, (
+        f"{required!r} is the core action set - must stay exposed for Gemma 4"
+    )

{drydock_cli-2.9.97 → drydock_cli-2.9.101}/tests/test_grammar.py

@@ -258,9 +258,32 @@ def _make_tool_manager_multi(specs: list[tuple[str, type]]) -> Any:
     return mgr
 
 
-def test_select_union_grammar_auto(monkeypatch):
+def test_select_union_grammar_auto_off_by_default(monkeypatch):
+    """v2.9.98: union grammar is OFF by default — must be explicitly
+    enabled via DRYDOCK_UNION_GRAMMAR_ENABLE=1. The 72-tool union let the
+    model pick whichever branch had the simplest schema and iterate
+    through them mechanically (task, exit_plan_mode, notebook_edit, …).
+    Forced-tool grammar (policy.py) is unaffected.
+    """
     monkeypatch.delenv("DRYDOCK_GRAMMAR_DISABLE", raising=False)
     monkeypatch.delenv("DRYDOCK_UNION_GRAMMAR_DISABLE", raising=False)
+    monkeypatch.delenv("DRYDOCK_UNION_GRAMMAR_ENABLE", raising=False)
+    specs = [("write_file", SimpleArgs), ("read_file", OptionalArgs)]
+    at = _make_available_tools(specs)
+    mgr = _make_tool_manager_multi(specs)
+    gbnf = select_union_grammar(
+        tool_choice="auto",
+        available_tools=at,
+        tool_manager=mgr,
+    )
+    assert gbnf is None  # off by default
+
+
+def test_select_union_grammar_auto_when_enabled(monkeypatch):
+    """When DRYDOCK_UNION_GRAMMAR_ENABLE=1, the union grammar engages."""
+    monkeypatch.delenv("DRYDOCK_GRAMMAR_DISABLE", raising=False)
+    monkeypatch.delenv("DRYDOCK_UNION_GRAMMAR_DISABLE", raising=False)
+    monkeypatch.setenv("DRYDOCK_UNION_GRAMMAR_ENABLE", "1")
     specs = [("write_file", SimpleArgs), ("read_file", OptionalArgs)]
     at = _make_available_tools(specs)
     mgr = _make_tool_manager_multi(specs)

drydock_cli-2.9.101/tests/test_loop_regressions_2026_06_06.py

@@ -0,0 +1,206 @@
+"""Behavioral regression tests for the loops user surfaced on 2026-06-06.
+
+Tonight's sequence of escalating loop fixes (v2.9.89 -> v2.9.99) each
+shipped, exposed the next layer, and got patched in turn. These tests
+pin the exact symptoms so the regressions can't return silently:
+
+  v2.9.89: write_file({}) and {path: None} acceptance       (schema fix)
+  v2.9.91: 18-readonly hard-stop firing in subagents        (subagent guard)
+  v2.9.93: parent dispatching identical task back-to-back   (preflight)
+  v2.9.95: notebook_edit on .py files                       (gemma4 disable)
+  v2.9.97: task tool spammed under union grammar             (task disable)
+  v2.9.98: exit_plan_mode -> "Already in implementation"     (gemma4 disable)
+           + union grammar disabled by default
+
+Each test exercises ONE specific symptom against the real AgentLoop or
+the real ToolManager so we get a yes/no signal on the architectural
+fix, not the surface phrasing.
+"""
+from __future__ import annotations
+
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+
+from drydock.core.agents.models import AgentType
+from drydock.core.tools.manager import ToolManager
+
+
+# ---------------------------------------------------------------------------
+# Helper: gemma4-configured ToolManager
+# ---------------------------------------------------------------------------
+
+def _gemma4_tool_manager() -> ToolManager:
+    cfg = MagicMock()
+    cfg.enabled_tools = None
+    cfg.disabled_tools = None
+    cfg.mcp_servers = []
+    cfg.tool_paths = []
+
+    class FakeModel:
+        name = "gemma-4-26b"
+        alias = "gemma4"
+
+    cfg.get_active_model = lambda: FakeModel()
+    return ToolManager(config_getter=lambda: cfg)
+
+
+# ===========================================================================
+# v2.9.89 — Schema fix: write_file required fields
+# ===========================================================================
+
+class TestSchemaTighteningWriteFile:
+    """Under union grammar, vacuous-default args (path="", content="")
+    must be promoted to required. Otherwise the grammar accepts
+    `{"name":"write_file","arguments":{}}` and the model emits garbage."""
+
+    def test_write_file_path_now_required_in_tightened_schema(self):
+        from drydock.core.llm.grammar.policy_union import _tighten_args_schema
+        from drydock.core.tools.builtins.write_file import WriteFileArgs
+
+        tight = _tighten_args_schema(WriteFileArgs.model_json_schema())
+        assert "path" in tight.get("required", [])
+        assert "content" in tight.get("required", [])
+
+    def test_overwrite_bool_default_stays_optional(self):
+        """Boolean default=True is a legit "use default" optional, must NOT
+        be forced required (would break callers that omit it)."""
+        from drydock.core.llm.grammar.policy_union import _tighten_args_schema
+        from drydock.core.tools.builtins.write_file import WriteFileArgs
+
+        tight = _tighten_args_schema(WriteFileArgs.model_json_schema())
+        assert "overwrite" not in tight.get("required", [])
+
+
+# ===========================================================================
+# v2.9.91 — Subagent hard-stop skip
+# ===========================================================================
+
+class TestSubagentSkipsAdaptiveBudget:
+    """The 18-readonly hard-stop must skip when the running AgentLoop is
+    a SUBAGENT — explore/diagnostic/planner/builder are read-mostly by
+    design and the parent owns the overall budget."""
+
+    def test_main_agent_still_triggers_hard_stop(self):
+        """Sanity: agents (not subagents) still hit the hard-stop."""
+        profile = SimpleNamespace(
+            agent_type=AgentType.AGENT,
+            name="default",
+        )
+        # The condition we patched is exactly:
+        # `self.agent_profile.agent_type != AgentType.SUBAGENT`
+        # When the profile is AGENT, the hard-stop is reachable.
+        assert profile.agent_type != AgentType.SUBAGENT
+
+    def test_subagent_skips_hard_stop(self):
+        for sub_name in ("explore", "diagnostic", "planner", "builder"):
+            profile = SimpleNamespace(
+                agent_type=AgentType.SUBAGENT,
+                name=sub_name,
+            )
+            # Under v2.9.91+ the hard-stop check guards against firing
+            # in subagent contexts. Pin that the type comparison short-
+            # circuits as designed.
+            assert profile.agent_type == AgentType.SUBAGENT, (
+                f"subagent {sub_name!r} must be flagged SUBAGENT so the "
+                f"hard-stop in agent_loop.py:~1185 skips."
+            )
+
+
+# ===========================================================================
+# v2.9.93 — Consecutive task call preflight (defense-in-depth even though
+# task is now disabled for Gemma 4 in v2.9.97).
+# ===========================================================================
+
+class TestConsecutiveTaskBlockPreflight:
+    """When a model emits `task` two turns in a row, the second call must
+    be blocked at preflight (drop from history + emit advisory). The
+    preflight is reachable only for non-Gemma models now that v2.9.97
+    yanks `task` from the Gemma 4 grammar, but it remains the safety
+    net for any future re-enable or non-Gemma backend.
+
+    The preflight code path is in agent_loop.py: hunt the marker comment
+    so a refactor that moves it surfaces here.
+    """
+
+    def test_preflight_marker_present_in_agent_loop(self):
+        """Regression guard: the consecutive-task preflight comment +
+        guard must exist in agent_loop.py. If someone refactors the
+        preflight chain and accidentally drops this block, this test
+        fires loud."""
+        text = open(
+            "/data3/drydock/drydock/core/agent_loop.py"
+        ).read()
+        assert "PREFLIGHT-TASK" in text, (
+            "consecutive-task preflight was removed from agent_loop.py"
+        )
+        assert "task tool blocked" in text, (
+            "consecutive-task advisory phrasing was changed; update test"
+        )
+
+
+# ===========================================================================
+# v2.9.95 / v2.9.97 / v2.9.98 — Tools that must be disabled for Gemma 4
+# ===========================================================================
+
+class TestGemma4DisableListCovers2026_06_06Loops:
+    """Every tool that drove a visible Gemma 4 loop session tonight
+    must be filtered out of the auto-discovered list. Parametrized in
+    test_gemma4_tool_filter.py too; these are duplicate explicit
+    assertions so a refactor that breaks the param suite still catches."""
+
+    def test_task_blocked(self):
+        names = set(_gemma4_tool_manager().available_tools.keys())
+        assert "task" not in names
+
+    def test_notebook_edit_blocked(self):
+        names = set(_gemma4_tool_manager().available_tools.keys())
+        assert "notebook_edit" not in names
+
+    def test_exit_plan_mode_blocked(self):
+        names = set(_gemma4_tool_manager().available_tools.keys())
+        assert "exit_plan_mode" not in names
+
+
+# ===========================================================================
+# v2.9.98 — Union grammar disabled by default
+# ===========================================================================
+
+class TestUnionGrammarDefaultOff:
+    """Union grammar (one branch per tool, ~13 vacuous-args branches)
+    must be OFF by default. The 72-tool union grammar made the model
+    iterate through whichever branch had the simplest schema. Default
+    is OFF; explicit opt-in via DRYDOCK_UNION_GRAMMAR_ENABLE=1."""
+
+    def test_default_off(self, monkeypatch):
+        monkeypatch.delenv("DRYDOCK_GRAMMAR_DISABLE", raising=False)
+        monkeypatch.delenv("DRYDOCK_UNION_GRAMMAR_DISABLE", raising=False)
+        monkeypatch.delenv("DRYDOCK_UNION_GRAMMAR_ENABLE", raising=False)
+        from drydock.core.llm.grammar.policy_union import select_union_grammar
+
+        mgr = MagicMock()
+        mgr.available_tools = {"write_file": object}
+        at = [MagicMock()]
+        at[0].function.name = "write_file"
+        gbnf = select_union_grammar(
+            tool_choice="auto",
+            available_tools=at,
+            tool_manager=mgr,
+        )
+        assert gbnf is None
+
+    def test_explicit_enable_engages(self, monkeypatch):
+        monkeypatch.delenv("DRYDOCK_GRAMMAR_DISABLE", raising=False)
+        monkeypatch.delenv("DRYDOCK_UNION_GRAMMAR_DISABLE", raising=False)
+        monkeypatch.setenv("DRYDOCK_UNION_GRAMMAR_ENABLE", "1")
+        from drydock.core.llm.grammar.policy_union import (
+            build_union_gbnf,
+            select_union_grammar,
+        )
+        from drydock.core.tools.builtins.write_file import WriteFileArgs
+
+        # Direct build path
+        gbnf = build_union_gbnf([("write_file", WriteFileArgs)])
+        assert gbnf is not None
+        assert "write_file" in gbnf or "root" in gbnf

{drydock_cli-2.9.97 → drydock_cli-2.9.101}/uv.lock

@@ -326,7 +326,7 @@ wheels = [
 
 [[package]]
 name = "drydock-cli"
-version = "2.9.89"
+version = "2.9.99"
 source = { editable = "." }
 dependencies = [
     { name = "agent-client-protocol" },