drydock-cli 2.9.115__tar.gz → 2.9.117__tar.gz

{drydock_cli-2.9.115 → drydock_cli-2.9.117}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.9.115
+Version: 2.9.117
 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-2.9.115 → drydock_cli-2.9.117}/drydock/core/agent_loop.py

@@ -1223,9 +1223,33 @@ class AgentLoop:
             PER_PROMPT_BUDGET_SEC = int(
                 getattr(self, "_admiral_per_prompt_budget_sec", 30 * 60)
             )
-            HARD_STOP_CALLS = int(
-                getattr(self, "_admiral_hard_stop_tool_calls", 100)
-            )
+            # 2026-06-07 (v2.9.116): raised from 100 -> 500. The lower cap
+            # was firing on legitimate complex investigations (operator's
+            # slide-reviewer session hit 100 while genuinely debugging a
+            # silent app failure — every tool call yielded new info, but
+            # the dumb count cap killed it). Keep it as a runaway safety
+            # net but pitched far above the "complex but real work"
+            # ceiling. The no-write streak (NO_WRITE_HARD_STOP, below)
+            # is the actual stuck-without-progress detector. Override
+            # via DRYDOCK_HARD_STOP_CALLS env or per-(model,task) admiral.
+            HARD_STOP_CALLS = int(os.environ.get(
+                "DRYDOCK_HARD_STOP_CALLS",
+                getattr(self, "_admiral_hard_stop_tool_calls", 500),
+            ))
+            # 2026-06-07 (v2.9.116): new no-write streak hard-stop.
+            # Fires when the model has done N consecutive readonly/neutral
+            # tool calls without committing a single write_file /
+            # search_replace / mechanical_rename. Catches "exploration
+            # without convergence" — the gap that HARD_STOP_CALLS=100 was
+            # plugging with an arbitrary count. Reset by any write tool
+            # or by a substantive plain-text response (model summarizing
+            # for the user counts as progress, just not file-system
+            # progress). Skip for SUBAGENT contexts because investigation
+            # subagents are legitimately read-mostly.
+            NO_WRITE_HARD_STOP = int(os.environ.get(
+                "DRYDOCK_NO_WRITE_HARD_STOP",
+                getattr(self, "_admiral_no_write_hard_stop", 80),
+            ))
             WRAP_UP_WARN_AT = int(getattr(self, "_admiral_wrap_up_warn_at",
                 int(os.environ.get("DRYDOCK_WRAP_UP_WARN_AT", 30))))
             STOP_NOW_WARN_AT = int(getattr(self, "_admiral_stop_now_warn_at",
@@ -1407,22 +1431,43 @@ class AgentLoop:
                         + (f" {_stop_suffix}" if _stop_suffix else "")
                     )
                 elif tool_turns >= HARD_STOP_CALLS:
-                    # Hard end-of-turn: synthesize a user-facing message
-                    # and stop. Was 50 but issue #9 showed "really
-                    # difficult" tasks legitimately need more than 50
-                    # tool calls. 100 preserves the runaway-loop safety
-                    # while giving complex builds room to finish.
+                    # v2.9.116: cap raised 100 -> 500. The no-write-streak
+                    # hard-stop below is the real "stuck without progress"
+                    # detector. This remains as runaway safety.
                     yield AssistantEvent(
                         content=(
                             f"\n\n[Drydock: stopped after {tool_turns} tool "
-                            "calls on a single request — too long without "
-                            "closing the turn. Returning control to the "
+                            "calls on a single request — runaway safety "
+                            "limit reached. Returning control to the "
                             "user. The work done so far is on disk; your "
                             "next prompt can review or continue.]\n"
                         ),
                         stopped_by_middleware=True,
                     )
                     return
+                # v2.9.116: no-write-streak hard-stop. Fires when the
+                # model has spent NO_WRITE_HARD_STOP turns reading and
+                # running diagnostics without committing a single edit.
+                # Skip for SUBAGENT contexts (explore/diagnostic/...) —
+                # they're read-mostly by design and the parent owns the
+                # overall budget. Same gate v2.9.91 added for the
+                # readonly-streak hard-stop.
+                no_write_streak = getattr(self, "_no_write_streak", 0)
+                if (no_write_streak >= NO_WRITE_HARD_STOP
+                        and self.agent_profile.agent_type
+                        != AgentType.SUBAGENT):
+                    yield AssistantEvent(
+                        content=(
+                            f"\n\n[Drydock: {no_write_streak} consecutive "
+                            "tool calls without any write/edit. Stuck in "
+                            "exploration. Returning control to the user. "
+                            "If you need to continue, your next prompt "
+                            "can pick up from where this trail off — "
+                            "every read result is still in history.]\n"
+                        ),
+                        stopped_by_middleware=True,
+                    )
+                    return
                 if tool_turns in (75, 125, 175):
                     self._inject_system_note(
                         f"You have used {tool_turns}/{MAX_TOOL_TURNS} tool calls. "
@@ -4244,6 +4289,12 @@ class AgentLoop:
             self._readonly_streak = 0
         if not hasattr(self, "_last_reflection_streak"):
             self._last_reflection_streak = 0
+        # v2.9.116: parallel counter — tool calls since the last write.
+        # Grows on readonly + neutral, resets only on write tools or a
+        # substantive assistant text response (resetting on text is
+        # handled in the main loop where assistant content is observed).
+        if not hasattr(self, "_no_write_streak"):
+            self._no_write_streak = 0
 
         # Reclassify bash → readonly when it's a read-shaped command.
         # Pull the command out of args; default to "" if missing.
@@ -4284,7 +4335,12 @@ class AgentLoop:
         if effective_kind == "write":
             self._readonly_streak = 0
             self._last_reflection_streak = 0
+            self._no_write_streak = 0  # v2.9.116: real write -> progress
             return
+        # v2.9.116: every readonly + neutral grows the no-write streak.
+        # Reset only on a real write (above) or substantive assistant
+        # text (handled in the main loop after we see the LLM response).
+        self._no_write_streak += 1
         if effective_kind == "readonly":
             # 2026-06-07 (v2.9.111/v2.9.114): Only count toward the streak
             # when this readonly call's RESULT was already seen in the

{drydock_cli-2.9.115 → drydock_cli-2.9.117}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.9.115"
+version = "2.9.117"
 description = "Local-first CLI coding agent — tested with Gemma 4 26B via llama.cpp"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-2.9.115 → drydock_cli-2.9.117}/tests/test_loop_regressions_2026_06_06.py

@@ -388,6 +388,63 @@ class TestReadonlyStreakOnlyCountsDoomLoops:
         assert "doom-loop signature is" in text or "v2.9.111" in text
 
 
+class TestNoWriteStreakHardStop:
+    """v2.9.116: replace the arbitrary 100-call hard stop with two
+    layered limits — HARD_STOP_CALLS raised to 500 as runaway safety,
+    and a new NO_WRITE_HARD_STOP at 80 consecutive readonly/neutral
+    tool calls without any write/edit. Operator's reasoning: real
+    work shouldn't get killed by a count cap, only real loops should
+    stop. The no-write streak is the actual "stuck without progress"
+    detector."""
+
+    def test_hard_stop_calls_raised_from_100_to_500(self):
+        """The arbitrary count cap that killed legit work."""
+        text = open(
+            "/data3/drydock/drydock/core/agent_loop.py"
+        ).read()
+        assert (
+            '"DRYDOCK_HARD_STOP_CALLS"' in text
+            and 'getattr(self, "_admiral_hard_stop_tool_calls", 500)' in text
+        ), "HARD_STOP_CALLS default should be 500 (env-configurable)"
+
+    def test_no_write_hard_stop_default_80(self):
+        text = open(
+            "/data3/drydock/drydock/core/agent_loop.py"
+        ).read()
+        assert "DRYDOCK_NO_WRITE_HARD_STOP" in text
+        assert '"_admiral_no_write_hard_stop", 80' in text
+
+    def test_no_write_streak_skips_subagent(self):
+        """The no-write streak hard-stop must skip subagents (explore /
+        diagnostic / planner / builder) — they're read-mostly by design.
+        Same gate v2.9.91 added for the readonly-streak hard-stop."""
+        text = open(
+            "/data3/drydock/drydock/core/agent_loop.py"
+        ).read()
+        # Find the no-write check block and verify it has the SUBAGENT
+        # type check
+        idx = text.find("no_write_streak >= NO_WRITE_HARD_STOP")
+        assert idx >= 0, "no-write hard-stop block missing"
+        # Within a reasonable window after, AgentType.SUBAGENT must
+        # appear (the skip-condition)
+        window = text[idx : idx + 400]
+        assert "AgentType.SUBAGENT" in window, (
+            "subagent skip missing from no-write hard-stop"
+        )
+
+    def test_write_resets_no_write_streak(self):
+        """Verify the reset path in _maybe_reflect: a write tool sets
+        _no_write_streak back to 0."""
+        text = open(
+            "/data3/drydock/drydock/core/agent_loop.py"
+        ).read()
+        # The reset is paired with the readonly-streak reset
+        idx = text.find('if effective_kind == "write":')
+        assert idx >= 0
+        window = text[idx : idx + 300]
+        assert "self._no_write_streak = 0" in window
+
+
 class TestContextWindowProbe:
     """v2.9.115: drydock probes the live llama.cpp backend's /props
     endpoint at AgentLoop init to discover the runtime n_ctx, then

{drydock_cli-2.9.115 → drydock_cli-2.9.117}/uv.lock

@@ -326,7 +326,7 @@ wheels = [
 
 [[package]]
 name = "drydock-cli"
-version = "2.9.114"
+version = "2.9.115"
 source = { editable = "." }
 dependencies = [
     { name = "agent-client-protocol" },