drydock-cli 2.9.34__tar.gz → 2.9.36__tar.gz

{drydock_cli-2.9.34 → drydock_cli-2.9.36}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.9.34
+Version: 2.9.36
 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-2.9.34 → drydock_cli-2.9.36}/drydock/cli/cli.py

@@ -322,6 +322,33 @@ def run_cli(args: argparse.Namespace) -> None:
         else:
             _t6 = _t5
 
+        # 2026-06-01: programmatic-mode escape hatch for -p with --output != text.
+        # The TUI path floods stdout with ANSI escape codes (19MB of TUI
+        # rendering per tbench trial — observed today). Harness consumers
+        # (terminal-bench, SWE-bench, harbor) need clean JSON per turn.
+        # run_programmatic() exists for this; route through it when
+        # --output json/streaming is explicitly requested.
+        if args.prompt is not None and getattr(args, "output", "text") != "text":
+            _phase("STARTUP_TOTAL_to_programmatic", _t0)
+            from drydock.core.types import OutputFormat
+            output_fmt = OutputFormat(args.output)
+            previous_messages = None
+            if loaded_session:
+                # loaded_session is (messages, session_id) — unpack just the messages
+                previous_messages = loaded_session[0] if loaded_session else None
+            final = run_programmatic(
+                config=config,
+                prompt=args.initial_prompt or stdin_prompt,
+                max_turns=getattr(args, "max_turns", None),
+                max_price=getattr(args, "max_price", None),
+                output_format=output_fmt,
+                previous_messages=previous_messages,
+                agent_name=initial_agent_name,
+            )
+            if final:
+                print(final)
+            return
+
         _phase("STARTUP_TOTAL_to_TUI_launch", _t0)
         run_textual_ui(
             agent_loop=agent_loop,

{drydock_cli-2.9.34 → drydock_cli-2.9.36}/drydock/core/agent_loop.py

@@ -2338,9 +2338,27 @@ class AgentLoop:
                 f"  - or end your turn with a text summary explaining "
                 f"what you intended to create."
             )
-        # For read-only tools include the full cached content so the model
-        # has the data it needs and doesn't retry just to see more output.
-        result_preview = last_result if is_readonly else last_result[:200]
+        # For read-only tools: embed the cached content so the model
+        # has the data, but CAP at 800 chars to stop the bait.
+        # Operator session 2026-06-01: read_file got skipped at count=6
+        # for a 40+ line renderer.py, and the FULL FILE content was
+        # echoed back in the skip-notice — every line of the source the
+        # model could pattern-match and emit as the next call's
+        # arguments. The previous logic was "include full result so
+        # the model has the data" but the model already has the data
+        # (it's been read 5+ times this session — earlier tool_result
+        # messages still carry it). The repeat-skip is meant to NUDGE,
+        # not re-provide.
+        if is_readonly:
+            if len(last_result) > 800:
+                result_preview = (
+                    last_result[:400]
+                    + f"\n  …[{len(last_result)-400} chars omitted on dedup re-display]"
+                )
+            else:
+                result_preview = last_result
+        else:
+            result_preview = last_result[:200]
         return (
             f"NOTE: this exact call to `{tool_name}` has been made "
             f"{count} times this session with identical arguments. "
@@ -5562,9 +5580,11 @@ class AgentLoop:
                     if _tc.function and _tc.function.name == "bash":
                         try:
                             _a = json.loads(_tc.function.arguments or "{}")
-                            _cmd = _a.get("command", "")
-                            if _cmd:
-                                _bash_cmds.append(_cmd)
+                            # Normalize empty/missing command to "" so the
+                            # consecutive-identical check fires for bash({})
+                            # loops (the `if _cmd:` guard was excluding them).
+                            _cmd = _a.get("command", "") or ""
+                            _bash_cmds.append(_cmd)
                         except (json.JSONDecodeError, AttributeError):
                             pass
             if _total_tc >= 20:

{drydock_cli-2.9.34 → drydock_cli-2.9.36}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.9.34"
+version = "2.9.36"
 description = "Local-first CLI coding agent — tested with Gemma 4 26B via llama.cpp"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-2.9.34 → drydock_cli-2.9.36}/tests/test_loop_detection.py

@@ -664,6 +664,66 @@ class TestBashEmptyCommand:
         )
 
 
+class TestBashEmptyCommandLoopStop:
+    """3 consecutive bash({}) calls must trigger FORCE_STOP via _bash_cmds.
+    Before the fix, `if _cmd:` excluded empty commands from _bash_cmds so
+    the consecutive-3-identical check never fired (319 failed interventions
+    in 24h observed in dispatch queue, 2026-06-02)."""
+
+    def _make_bash_tc(self, cmd: str | None = None) -> object:
+        """Build a minimal tool-call stub for bash with the given command."""
+        import types
+        fn = types.SimpleNamespace(
+            name="bash",
+            arguments=f'{{"command": "{cmd}"}}' if cmd is not None else "{}",
+        )
+        return types.SimpleNamespace(function=fn)
+
+    def _make_assistant_msg(self, cmd: str | None) -> object:
+        import types
+        from drydock.core.types import Role
+        return types.SimpleNamespace(
+            role=Role.assistant,
+            tool_calls=[self._make_bash_tc(cmd)],
+            content=None,
+        )
+
+    def test_empty_bash_included_in_bash_cmds(self):
+        """_bash_cmds must include '' for bash({}) so consecutive check fires."""
+        import json
+        # Replicate the extraction logic from agent_loop.py
+        bash_cmds: list[str] = []
+        for cmd_raw in [{}, {"command": ""}, {"command": None}]:
+            _a = cmd_raw
+            _cmd = _a.get("command", "") or ""  # the fixed version
+            bash_cmds.append(_cmd)
+        # All three must be "" so 3-consecutive fires
+        assert bash_cmds == ["", "", ""], (
+            f"Empty bash calls not tracked as '': {bash_cmds}"
+        )
+
+    def test_three_consecutive_empty_bash_signals_loop(self):
+        """With the fix, ['']*3 consecutive in _bash_cmds triggers the check."""
+        cmds = ["", "", ""]
+        # Simulate the check
+        triggered = len(cmds) >= 3 and cmds[0] == cmds[1] == cmds[2]
+        assert triggered, "Consecutive-3-empty check should fire"
+
+    def test_old_guard_would_miss_empty(self):
+        """Without the fix (if _cmd guard), empty commands were not appended."""
+        bash_cmds_old: list[str] = []
+        for cmd in ["", "", ""]:
+            if cmd:  # the old guard — intentionally wrong
+                bash_cmds_old.append(cmd)
+        # Old code: list stays empty, check can't fire
+        triggered_old = (
+            len(bash_cmds_old) >= 3
+            and bash_cmds_old[0:3] == [bash_cmds_old[0]] * 3
+            if len(bash_cmds_old) >= 3 else False
+        )
+        assert not triggered_old, "Old guard should NOT trigger (demonstrates the bug)"
+
+
 # ============================================================================
 # Mid-turn user injection (Claude Code "type while busy" feature)
 # ============================================================================