drydock-cli 2.9.48__tar.gz → 2.9.51__tar.gz

{drydock_cli-2.9.48 → drydock_cli-2.9.51}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.9.48
+Version: 2.9.51
 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-2.9.48 → drydock_cli-2.9.51}/drydock/admiral/llm_analyzer.py

@@ -65,7 +65,7 @@ really going wrong and propose ONE short directive to unstick the agent.
 
 Detector that fired: {code}
 Initial directive Admiral was about to send: "{fallback}"
-
+{loop_constraint}
 Recent conversation (last {n_turns} turns):
 <conversation>
 {context}
@@ -108,10 +108,22 @@ async def analyze(agent_loop: AgentLoop, finding: Finding) -> str | None:
     """
     try:
         context = _render_turns(agent_loop.messages, n_turns=12)
+        # If the loop is on a bash command, add an explicit constraint so
+        # the LLM doesn't suggest running that same command again.
+        loop_constraint = ""
+        if finding.code.startswith("loop:bash::"):
+            cmd_snippet = finding.code[len("loop:bash::"):][:120]
+            loop_constraint = (
+                f"\nCRITICAL CONSTRAINT: The repeated command is `bash` with "
+                f"args `{cmd_snippet}`. Your directive MUST NOT suggest running "
+                f"this same command again — the model already has its output. "
+                f"Direct the agent to read/edit source code instead.\n"
+            )
         prompt = _ANALYZER_PROMPT.format(
             rubric=_DIRECTIVES_RUBRIC,
             code=finding.code,
             fallback=finding.directive[:200],
+            loop_constraint=loop_constraint,
             n_turns=12,
             context=context,
         )

{drydock_cli-2.9.48 → drydock_cli-2.9.51}/drydock/core/agent_loop.py

@@ -1642,15 +1642,30 @@ class AgentLoop:
                             if fn in ("write_file", "search_replace"):
                                 writes_so_far += 1
                     zero_writes = writes_so_far == 0
-                    # Cap raised from 2 → 5 so we have headroom to nudge
-                    # subsequent text-only attempts. tool_turns gate raised
-                    # 5 → 20 because some "browse then quit" patterns happen
-                    # AFTER 6-15 explorations (e.g. mailman: bash:21, read:6).
-                    if is_programmatic and nudges < 5 and tool_turns < 20:
+                    # Cap raised 5 → 10 nudges per session. Removed tool_turns
+                    # gate entirely — the 2026-06-03 v2948 batch showed that
+                    # most "engaged but failed" trials (bucket D, 26/50 fails)
+                    # blew past tool_turns=20 with extensive engagement and
+                    # THEN emitted a "I have completed X" text. Examples:
+                    # chess-best-move 107 msgs, raman-fitting 156 msgs,
+                    # gpt2-codegolf 126 msgs — all over the previous gate.
+                    # The verifier never ran for any of them. Nudging late
+                    # forces them to run the verifier and iterate.
+                    #
+                    # Also count `verify` tool calls. If the model emitted a
+                    # `verify` call recently AND it returned passed=True, do
+                    # NOT nudge — the model genuinely passed its check.
+                    verify_count = 0
+                    last_verify_passed = False
+                    for msg in self.messages:
+                        if msg.role == Role.tool and (msg.name == "verify"):
+                            verify_count += 1
+                            last_verify_passed = "passed: True" in (msg.content or "")
+                    if is_programmatic and nudges < 10 and not last_verify_passed:
                         self._premature_exit_nudges = nudges + 1
                         if zero_writes:
                             note = (
-                                f"⚠ PREMATURE EXIT BLOCK #{nudges + 1}/5. "
+                                f"⚠ PREMATURE EXIT BLOCK #{nudges + 1}/10. "
                                 "You have made ZERO write_file or search_replace "
                                 "calls. The task explicitly requires creating, "
                                 "modifying, or producing output file(s). Reading "
@@ -1664,23 +1679,28 @@ class AgentLoop:
                             )
                         else:
                             note = (
-                                f"⚠ PREMATURE EXIT BLOCK #{nudges + 1}/5. "
+                                f"⚠ PREMATURE EXIT BLOCK #{nudges + 1}/10. "
                                 f"You emitted a text-only response after "
                                 f"{tool_turns} tool call(s) and "
-                                f"{writes_so_far} write(s). The verifier hasn't "
-                                "passed — the task is NOT done. Emit the next "
-                                "concrete tool call NOW. Run the verifier "
-                                "command, fix whatever it complains about, "
-                                "iterate. Don't stop until pytest / the "
-                                "specified verify command returns 0."
+                                f"{writes_so_far} write(s). The verifier has "
+                                f"NOT passed (verify_calls={verify_count}, "
+                                "last_pass=False). Saying 'I have completed' "
+                                "in text doesn't count — RUN THE VERIFIER NOW. "
+                                "Use the `verify` tool, or `bash` to run "
+                                "pytest, the specified test command, or your "
+                                "best guess at what the verifier checks. If "
+                                "the verifier fails, READ the failure and FIX "
+                                "the source, then run it again. Don't stop "
+                                "until verify returns passed=True."
                             )
                         self._inject_system_note(note)
                         should_break_loop = False
                         logger.warning(
                             "[PREMATURE-EXIT] text-only response at turn %d "
-                            "(writes=%d) in programmatic mode — injected "
-                            "continuation nudge (#%d/5), continuing loop",
-                            tool_turns, writes_so_far, self._premature_exit_nudges,
+                            "(writes=%d, verifies=%d) in programmatic mode — "
+                            "injected continuation nudge (#%d/10), continuing loop",
+                            tool_turns, writes_so_far, verify_count,
+                            self._premature_exit_nudges,
                         )
 
                 # 2026-05-25: AUTO-GOAL loop. If a rename goal was

{drydock_cli-2.9.48 → drydock_cli-2.9.51}/drydock/core/llm/format.py

@@ -361,6 +361,12 @@ class APIToolFormatHandler:
             # server) and `read_mcp_resource` after running out of obvious
             # exploration tools. Suppress to break the empty_after_tool stall.
             "lsp", "lsp_definition", "lsp_references",
+            # Observed 2026-06-03 gauntlet sessions: model hallucinates rename/refactor
+            # tools, then falls back to empty bash({}) calls when they fail — triggering
+            # bash loop detector. Suppress with explicit search_replace redirect so the
+            # model uses the correct tool rather than looping on empty bash.
+            "mechanical_rename", "rename_symbol", "rename_identifier",
+            "refactor_rename", "rename_in_file", "symbol_rename",
         }
         # Subset of _IGNORE_TOOLS that are retrieval-flavored hallucinations.
         # When `retrieve` is registered, redirect the model there instead of
@@ -370,6 +376,12 @@ class APIToolFormatHandler:
             "ralph_repo_index", "repo_index", "index_repo",
             "ralph_file_summary", "file_summary", "repo_summary",
         }
+        # Rename/refactor hallucinations — redirect to search_replace with
+        # replace_all=True, which is the correct drydock primitive for renaming.
+        _RENAME_HALLUCINATIONS = {
+            "mechanical_rename", "rename_symbol", "rename_identifier",
+            "refactor_rename", "rename_in_file", "symbol_rename",
+        }
 
         suppressed_failures = []
         for parsed_call in parsed.tool_calls:
@@ -379,7 +391,15 @@ class APIToolFormatHandler:
                 # history well-formed (prevents empty_after_tool loops) but
                 # route to suppressed_failures so the TUI doesn't show an error.
                 if parsed_call.tool_name in _IGNORE_TOOLS:
-                    if parsed_call.tool_name in _RETRIEVAL_HALLUCINATIONS:
+                    if parsed_call.tool_name in _RENAME_HALLUCINATIONS:
+                        redirect = (
+                            f"'{parsed_call.tool_name}' does not exist. "
+                            f"To rename a symbol across a file use: "
+                            f"`search_replace(file_path='...', old_string='OldName', "
+                            f"new_string='NewName', replace_all=True)`. "
+                            f"Call search_replace NOW with the exact old and new names."
+                        )
+                    elif parsed_call.tool_name in _RETRIEVAL_HALLUCINATIONS:
                         # These are file-listing/indexing hallucinations, NOT semantic
                         # search — redirecting to `retrieve` confuses the model (it doesn't
                         # know what query to write for a directory listing, stalls empty).

{drydock_cli-2.9.48 → drydock_cli-2.9.51}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.9.48"
+version = "2.9.51"
 description = "Local-first CLI coding agent — tested with Gemma 4 26B via llama.cpp"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-2.9.48 → drydock_cli-2.9.51}/tests/test_hallucinated_tool_suppression.py

@@ -52,6 +52,13 @@ HALLUCINATED_TOOLS = [
     "lsp",
     "lsp_definition",
     "lsp_references",
+    # 2026-06-03 gauntlet: rename/refactor hallucinations that caused bash loop fallback.
+    "mechanical_rename",
+    "rename_symbol",
+    "rename_identifier",
+    "refactor_rename",
+    "rename_in_file",
+    "symbol_rename",
 ]
 
 

{drydock_cli-2.9.48 → drydock_cli-2.9.51}/uv.lock

@@ -326,7 +326,7 @@ wheels = [
 
 [[package]]
 name = "drydock-cli"
-version = "2.9.41"
+version = "2.9.49"
 source = { editable = "." }
 dependencies = [
     { name = "agent-client-protocol" },