drydock-cli 2.5.5__tar.gz → 2.5.7__tar.gz

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/CLAUDE.md

@@ -9,9 +9,12 @@ Drydock is a local CLI coding agent (fork of mistral-vibe, Apache 2.0).
 - **Hardware:** 2x RTX 4060 Ti 16GB, Gemma 4 26B MoE (A4B) via vLLM Docker at localhost:8000
 - **Server:** remus (Ubuntu 22.04, user: bobef)
 - **Active model:** Gemma 4 26B-A4B-it-AWQ-4bit (only 4B active params, ~70 tok/s)
-- **TUI results:** 88/90 PRD projects pass (98%) through the real TUI
-- **SWE-bench:** 70% file match (v3 baseline), 58% (v3 tuned — tighter prompt hurt)
-- **Priority:** TUI experience first, SWE-bench second
+- **TUI build tests:** 303/320 pass --help (94%) — BUT --help IS MEANINGLESS
+- **TUI real functional tests:** 35/63 pass (56%) — the honest number
+- **SWE-bench:** 70% file match (v3 baseline)
+- **Priority:** TUI experience first. Fix drydock bugs, don't simplify PRDs.
+- **370 PRDs** at /data3/drydock_test_projects/ — the benchmark suite
+- **Current version:** v2.5.6 (PyPI + GitHub + user's env)
 
 ## Build & Test
 
@@ -279,21 +282,22 @@ v3 is a from-scratch rewrite using nano-claude-code as foundation. 4 core files,
 
 ### TUI Test Harness
 - **Script:** `scripts/tui_test.py` — pexpect-based, drives the REAL TUI binary
-- **Auto-approves** tool permission prompts
-- **Detects:** write loops, API errors, tool call counts, file creation, --help pass/fail
-- **100 PRDs:** at `/data3/drydock_test_projects/` (88/90 pass = 98%)
-
-### PRD Format (what works)
-- Short (under 30 lines)
-- 4-6 files max
-- Clear package name and file list
-- Test command: `python3 -m package_name --help`
-- Stdlib only (no external deps)
-
-### What breaks PRDs
-- Too many files (8+) — model runs out of turns
-- Complex parsers (custom YAML/XML from scratch)
-- Features requiring external APIs or databases
+- **Real test runner:** `run_real_tests.sh` — functional testing, not just --help
+- **Results:** `real_test_results.txt` — honest pass/fail with functional verification
+- **370 PRDs** at `/data3/drydock_test_projects/` covering CLI tools, integration, edge cases, best practices
+
+### Test Levels
+1. **Build test (--help):** package runs — MEANINGLESS alone
+2. **Functional test:** PRD test cases verified with correct output — THE REAL TEST
+3. **Acceptance test:** specific expected outputs verified
+4. **Quality test:** code quality checks (docstrings, no eval, error handling)
+
+### Critical Testing Rules
+- PRD failures = DryDock bugs. Fix the agent, NOT the PRD.
+- --help passing proves nothing about code quality
+- Safety mechanisms must be ADVISORY not BLOCKING
+- NEVER add circuit breakers that prevent legitimate work
+- The model should be able to answer questions without being forced to edit files
 
 ## When Compacting
 

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.5.5
+Version: 2.5.7
 Summary: Drydock — nautical CLI coding agent
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/drydock/core/agent_loop.py

@@ -790,81 +790,21 @@ class AgentLoop:
                         continue
                     # After 3 empty responses, let it exit
 
-                # If model gives text without tool calls and hasn't edited anything,
-                # nudge it to make an edit instead of just describing what to do.
-                # Never let the agent exit without at least attempting an edit.
-                if should_break_loop and not has_made_edit and tool_turns >= 1:
-                    text_without_action += 1
-                    # Always continue — don't let the agent exit without editing
-                    should_break_loop = False
-
-                    # Detect model claiming task is done without having edited anything
-                    last_text_lower = (last_message.content or "").lower()
-                    task_complete_phrases = [
-                        "task completed", "task complete", "task is complete",
-                        "task is done", "task has been completed",
-                        "changes have been applied", "fix has been applied",
-                        "the fix is complete", "issue is resolved",
-                    ]
-                    claims_complete = any(p in last_text_lower for p in task_complete_phrases)
-
-                    if claims_complete:
-                        nudge_text = (
-                            "You said the task is complete but NO code was actually changed. "
-                            "Your search_replace call either failed or was never made. "
-                            "Use read_file to read the target file, then use search_replace "
-                            "to make your edit. Do NOT say the task is done until search_replace succeeds."
-                        )
-                    elif text_without_action == 1:
-                        last_text = (last_message.content or "").lower()
-                        # Detect model asking for confirmation instead of acting
-                        is_asking = any(kw in last_text for kw in (
-                            "standing by", "waiting for", "please provide",
-                            "what would you like", "how can i help",
-                            "what should i", "do you want me to",
-                            "ready to begin", "i will start",
-                            "shall i", "would you like me to",
-                            "let me know", "please confirm",
-                            "i am ready", "ready to proceed",
-                        ))
-                        if is_asking:
-                            nudge_text = (
-                                "DO NOT ask for confirmation. Act NOW. "
-                                "If there is a PRD.md, implement it with write_file. "
-                                "If there is code, read it with read_file and fix bugs with search_replace. "
-                                "Call a tool on your next response."
-                            )
-                        elif "TARGET:" in last_text.upper() or "FILE:" in last_text.upper():
-                            nudge_text = (
-                                "Good — you identified the target. Now use read_file then search_replace."
-                            )
-                        else:
-                            nudge_text = (
-                                "You responded with text but did not call any tools. "
-                                "Call write_file, read_file, or search_replace NOW."
-                            )
-                    elif text_without_action == 2:
-                        nudge_text = (
-                            "STOP describing the fix. You MUST call search_replace on your NEXT response. "
-                            "No more text-only responses. Act NOW."
-                        )
-                    elif text_without_action <= 5:
-                        nudge_text = (
-                            f"WARNING ({text_without_action} text responses without editing): "
-                            "You MUST call search_replace or read_file immediately. "
-                            "Make your best fix attempt RIGHT NOW even if you are unsure. "
-                            "Pick the most likely file and function, and make a minimal edit."
-                        )
-                    else:
-                        # Keep nudging — never give up
-                        nudge_text = (
-                            "You MUST call a tool. Use grep to find the file, "
-                            "read_file to read it, then search_replace to fix."
+                # Text response without edits — this is FINE for questions,
+                # explanations, running code, etc. Only nudge if the model
+                # is clearly asking for permission instead of acting.
+                if should_break_loop and not has_made_edit:
+                    last_text = (last_message.content or "").lower()
+                    is_asking = any(kw in last_text for kw in (
+                        "shall i", "would you like me to",
+                        "ready to begin", "please confirm",
+                    ))
+                    if is_asking and tool_turns == 0:
+                        # Model asked without doing anything — nudge once
+                        should_break_loop = False
+                        self._inject_system_note(
+                            "Act immediately. Do not ask for confirmation."
                         )
-                    if nudge_text:
-                        # Use _inject_system_note (appends to last tool/user message)
-                        # User messages cause 'tool after user' ordering violations
-                        self._inject_system_note(nudge_text)
                         logger.info("Model gave text without editing — nudging (attempt %d)", text_without_action)
 
                 # If model has been investigating for too long without making an edit,

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/drydock/core/session_checker.py

@@ -128,17 +128,20 @@ def check_session(messages: list, has_made_edit: bool = False) -> list[SessionIs
             f"{api_errors} API errors occurred during the session"
         ))
 
-    # 7. No edits made despite having a task
-    if not has_made_edit and len(messages) > 5:
-        has_user_task = any(
-            hasattr(msg, 'role') and str(getattr(msg, 'role', '')) == 'user'
-            and len(str(getattr(msg, 'content', '') or '')) > 20
-            for msg in messages
+    # 7. No edits — only flag if the user explicitly asked to build/fix/create something
+    # Questions, exploration, and running code don't need edits
+    if not has_made_edit and len(messages) > 10:
+        user_msgs = [str(getattr(m, 'content', '') or '').lower() for m in messages
+                     if hasattr(m, 'role') and str(getattr(m, 'role', '')) == 'user']
+        build_words = ["build", "create", "fix", "implement", "write", "add", "modify", "update", "edit"]
+        user_asked_for_edit = any(
+            any(w in msg for w in build_words)
+            for msg in user_msgs
         )
-        if has_user_task:
+        if user_asked_for_edit:
             issues.append(SessionIssue(
-                "warning", "no_edits",
-                "Session ended without making any file changes"
+                "info", "no_edits",
+                "No file changes were made — the model may not have completed the task"
             ))
 
     return issues

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/drydock/core/tools/builtins/search_replace.py

@@ -166,6 +166,17 @@ class SearchReplace(
 
             await self._write_file(file_path, modified_content)
 
+            # Auto-verify syntax for Python files
+            if file_path.suffix == ".py":
+                try:
+                    import ast
+                    ast.parse(modified_content)
+                except SyntaxError as e:
+                    block_result.warnings.append(
+                        f"⚠ SYNTAX ERROR after edit at line {e.lineno}: {e.msg}. "
+                        f"Fix this before continuing."
+                    )
+
         yield SearchReplaceResult(
             file=str(file_path),
             blocks_applied=block_result.applied,
@@ -188,7 +199,6 @@ class SearchReplace(
         # Try to extract file_path from content if missing
         if not file_path_str and content:
             # Look for file path in the content (common when model puts everything in content)
-            import re
             path_match = re.search(r'(?:file[_\s]?path|path|file)[\s:="\']+([^\s"\']+\.py)', content[:200], re.IGNORECASE)
             if path_match:
                 file_path_str = path_match.group(1)

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/drydock/core/tools/builtins/write_file.py

@@ -92,12 +92,28 @@ class WriteFile(
 
         await self._write_file(args, file_path)
 
-        yield WriteFileResult(
+        # Auto-verify syntax for Python files
+        syntax_warning = ""
+        if file_path.suffix == ".py":
+            try:
+                import ast
+                ast.parse(args.content)
+            except SyntaxError as e:
+                syntax_warning = (
+                    f"\n\n⚠ SYNTAX ERROR in {file_path.name} line {e.lineno}: {e.msg}"
+                    f"\n  {e.text.rstrip() if e.text else ''}"
+                    f"\n  Fix this before moving to the next file."
+                )
+
+        result = WriteFileResult(
             path=str(file_path),
             bytes_written=content_bytes,
             file_existed=file_existed,
             content=args.content,
         )
+        if syntax_warning:
+            result.content = syntax_warning  # Override content with the warning
+        yield result
 
     _BINARY_EXTENSIONS = frozenset({
         ".pptx", ".xlsx", ".docx", ".pdf", ".zip", ".tar", ".gz", ".bz2",

{drydock_cli-2.5.5 → drydock_cli-2.5.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.5.5"
+version = "2.5.7"
 description = "Drydock — nautical CLI coding agent"
 readme = "README.md"
 requires-python = ">=3.12"