drydock-cli 3.0.31__tar.gz → 3.0.32__tar.gz

{drydock_cli-3.0.31/drydock_cli.egg-info → drydock_cli-3.0.32}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.31
+Version: 3.0.32
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0

{drydock_cli-3.0.31 → drydock_cli-3.0.32}/drydock/__init__.py

@@ -7,4 +7,4 @@ try:
     __version__ = _version("drydock-cli")
 except PackageNotFoundError:
     # Running from source before install (e.g. PYTHONPATH dev runs).
-    __version__ = "3.0.31"
+    __version__ = "3.0.32"

{drydock_cli-3.0.31 → drydock_cli-3.0.32}/drydock/agent.py

@@ -115,13 +115,15 @@ def run(
     leaked_call_retries = 0
     plan_continue_nudges = 0  # consecutive "you stopped mid-plan" nudges
     empty_response_nudges = 0  # consecutive "you returned nothing" nudges
-    # Safety valve for a degenerate loop: the SAME tool call FAILING over and
-    # over (a weak model that ignores the advisory loop-note — seen failing a
-    # Write 160× on a too-large file). Real iterative fixing changes the args,
-    # so the streak resets; only byte-identical repeated FAILURES trip it.
-    identical_fail_streak = 0
-    last_fail_sig = None
-    IDENTICAL_FAIL_CAP = 8
+    # Safety valve for a degenerate loop: the SAME tool call run over and over
+    # with the SAME result — success OR failure (seen failing a Write 160×, and
+    # re-running an identical passing `pytest` 92× for 25 min). The advisory
+    # loop-note is ignored by weak models, so after a cap we end the turn. Real
+    # iterative fixing changes the args, and polling yields a changing result —
+    # both reset the streak, so only a truly pointless loop trips it.
+    identical_repeat_streak = 0
+    last_call_sig = None
+    IDENTICAL_REPEAT_CAP = 8
     run_iteration = 0  # stream calls within THIS run() (resets per user message)
     loop_tracker = LoopTracker()
 
@@ -295,16 +297,15 @@ def run(
                 )
             else:
                 result = execute(tc["name"], tc["input"], config)
-            # Track byte-identical repeated FAILURES (checked before annotate
-            # prepends its advisory note) for the safety valve below.
-            failed = result.lstrip().startswith(("Error", "REFUSED"))
-            sig = (tc["name"], str(tc["input"]))
-            if failed and sig == last_fail_sig:
-                identical_fail_streak += 1
-            elif failed:
-                identical_fail_streak, last_fail_sig = 1, sig
+            # Track consecutive byte-identical calls — same name, args AND raw
+            # result (captured before annotate prepends its note, which changes
+            # each call) — for the safety valve below. A differing result
+            # (polling) or differing args (real iteration) resets the streak.
+            sig = (tc["name"], str(tc["input"]), result)
+            if sig == last_call_sig:
+                identical_repeat_streak += 1
             else:
-                identical_fail_streak, last_fail_sig = 0, None
+                identical_repeat_streak, last_call_sig = 1, sig
             # Guide (never block) on exact-repeat tool calls: prepend an
             # advisory note when the same call is made again.
             result = loop_tracker.annotate(tc["name"], tc["input"], result)
@@ -319,16 +320,15 @@ def run(
                 "content": result,
             })
 
-        # Safety valve: the same call has FAILED identically too many times —
-        # repeating it won't help. End the turn and hand control back with a
-        # clear note rather than burning turns toward MAX_TOOL_TURNS.
-        if identical_fail_streak >= IDENTICAL_FAIL_CAP:
+        # Safety valve: the same call has run identically (same args AND result)
+        # too many times in a row — repeating it changes nothing. End the turn
+        # and hand control back rather than burning turns toward MAX_TOOL_TURNS.
+        if identical_repeat_streak >= IDENTICAL_REPEAT_CAP:
             yield TextChunk(
-                f"\n[Stopped: the same {last_fail_sig[0]} call failed "
-                f"{identical_fail_streak}× in a row — repeating it won't change "
-                "the result. If it's a Write, the file is likely too large for "
-                "one response; write it in smaller pieces. Control is back to "
-                "you.]\n"
+                f"\n[Stopped: the same {last_call_sig[0]} call ran "
+                f"{identical_repeat_streak}× in a row with the same result — "
+                "repeating it changes nothing. Control is back to you; tell me "
+                "how you'd like to proceed.]\n"
             )
             break
 

{drydock_cli-3.0.31 → drydock_cli-3.0.32/drydock_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.31
+Version: 3.0.32
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0

{drydock_cli-3.0.31 → drydock_cli-3.0.32}/pyproject.toml

@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
 # PyPI distribution name is drydock-cli (the established install name, continued
 # from the retired v2 fork); the import package + CLI command stay `drydock`.
 name = "drydock-cli"
-version = "3.0.31"
+version = "3.0.32"
 description = "Drydock — a local, provider-agnostic terminal coding agent for local LLMs"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-3.0.31 → drydock_cli-3.0.32}/tests/test_failure_loop.py

@@ -41,3 +41,37 @@ def test_changing_args_do_not_trip_valve(monkeypatch):
     list(run("go", st, {"model": "m", "cwd": tempfile.mkdtemp()}, "sys"))
     assert any(m.get("role") == "assistant" and m.get("content") == "done"
                for m in st.messages)
+
+
+def test_identical_SUCCESS_streak_ends_turn(monkeypatch):
+    # The code-red case: a SUCCESSFUL command re-run identically (same args AND
+    # same output) — a passing `pytest` 92×. Must also stop, not just failures.
+    import tempfile
+    calls = {"n": 0}
+
+    def counting(**kw):
+        calls["n"] += 1
+        return iter([AssistantTurn(
+            "", [{"id": str(calls["n"]), "name": "Bash",
+                  "input": {"command": "echo hi"}}], 1, 1)])
+
+    monkeypatch.setattr(agent_mod, "stream", counting)
+    st = AgentState()
+    list(run("go", st, {"model": "m", "cwd": tempfile.mkdtemp()}, "sys"))
+    assert calls["n"] <= 12  # identical success → stopped at the cap, not 200
+
+
+def test_polling_with_changing_result_not_tripped(monkeypatch):
+    # Same command, DIFFERENT result each time (a poll) must NOT be stopped —
+    # the result changes, so the streak resets. Model finishes on its own.
+    import tempfile
+    seq = []
+    for i in range(3):
+        seq.append(AssistantTurn(
+            "", [{"id": str(i), "name": "Bash",
+                  "input": {"command": f"echo {i}"}}], 1, 1))  # different output
+    seq.append(AssistantTurn("done", [], 1, 1))
+    monkeypatch.setattr(agent_mod, "stream", lambda **kw: iter([seq.pop(0)]))
+    st = AgentState()
+    list(run("go", st, {"model": "m", "cwd": tempfile.mkdtemp()}, "sys"))
+    assert any(m.get("content") == "done" for m in st.messages)