drydock-cli 1.5.0__tar.gz → 1.5.2__tar.gz

{drydock_cli-1.5.0 → drydock_cli-1.5.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 1.5.0
+Version: 1.5.2
 Summary: Drydock — nautical CLI coding agent
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-1.5.0 → drydock_cli-1.5.2}/drydock/core/agent_loop.py

@@ -473,6 +473,13 @@ class AgentLoop:
 
         yield UserMessageEvent(content=user_msg, message_id=user_message.message_id)
 
+        # === AUTO-DELEGATION ===
+        # Inject context automatically so the model doesn't have to choose to delegate.
+        # This replaces relying on the model to call task()/invoke_skill().
+        auto_context = self._build_auto_context(user_msg)
+        if auto_context:
+            self._inject_system_note(auto_context)
+
         try:
             should_break_loop = False
             tool_turns = 0
@@ -627,6 +634,41 @@ class AgentLoop:
                         f"Consider stopping to verify your approach."
                     )
 
+                # Post-edit bash loop: model made edits but now loops on bash
+                # trying to test code that keeps failing
+                if has_made_edit and bash_count >= 3:
+                    # Count consecutive bash calls at end of tool history
+                    consecutive_bash = 0
+                    for tc_name in reversed(self._recent_tool_names()):
+                        if tc_name in ("bash", "run_command"):
+                            consecutive_bash += 1
+                        else:
+                            break
+                    if consecutive_bash >= 4:
+                        self._inject_system_note(
+                            "You have run 4+ bash commands in a row after editing files. "
+                            "Each one failed or gave the same result. STOP running bash. "
+                            "Read the error message carefully, then use search_replace "
+                            "to fix the actual code. Running the same command again will "
+                            "not give a different result."
+                        )
+
+                # Detect bash file creation (touch, echo >, cat <<)
+                for msg in reversed(self.messages[-3:]):
+                    if msg.role == Role.assistant and msg.tool_calls:
+                        for tc in msg.tool_calls:
+                            if tc.function and tc.function.name in ("bash", "run_command"):
+                                cmd = ""
+                                try:
+                                    cmd = json.loads(tc.function.arguments or "{}").get("command", "")
+                                except (json.JSONDecodeError, AttributeError):
+                                    pass
+                                if any(kw in cmd for kw in ("touch ", "echo ", "cat <<", "printf ", "> ")):
+                                    self._inject_system_note(
+                                        f"Do NOT use bash to create files. Use write_file instead. "
+                                        f"bash touch/echo/cat does not create proper code files."
+                                    )
+
                 # Bash abuse detection: model uses bash instead of proper tools
                 if not has_made_edit and bash_count >= 3:
                     if bash_count == 3:
@@ -1203,17 +1245,33 @@ class AgentLoop:
                     "Add more surrounding context lines to old_str to make it unique.]"
                 )
 
-            # RECOVERY: After bash failure with traceback, extract file/line and nudge
+            # RECOVERY: Relative import error — tell model to use absolute imports or -m
+            if tool_call.tool_name in ("bash", "run_command") and "relative import with no known parent" in str(exc):
+                self._inject_system_note(
+                    "The error 'relative import with no known parent package' means you are "
+                    "running a package file directly (python3 pkg/file.py). Fix: either "
+                    "(1) change 'from .module import X' to 'from pkg.module import X' (absolute imports), or "
+                    "(2) run with 'python3 -m pkg' instead of 'python3 pkg/file.py'. "
+                    "Use search_replace to change the imports to absolute imports NOW."
+                )
+
+            # RECOVERY: After bash failure with traceback, extract file/line and
+            # inject a STRONG system note (not just error text) to force read→fix
             if tool_call.tool_name in ("bash", "run_command") and "Traceback" in str(exc):
                 import re
-                # Extract last File "path", line N from traceback
                 tb_matches = re.findall(r'File "([^"]+)", line (\d+)', str(exc))
                 if tb_matches:
                     tb_file, tb_line = tb_matches[-1]
-                    if not tb_file.startswith("/home") and not "site-packages" in tb_file:
+                    if not tb_file.startswith("/home") and "site-packages" not in tb_file:
                         error_msg += (
-                            f"\n\n[NEXT STEP: Read the source at {tb_file} around line {tb_line} "
-                            f"with read_file to understand the bug, then fix it with search_replace.]"
+                            f"\n\n[NEXT STEP: Read {tb_file} around line {tb_line} "
+                            f"with read_file, then fix it with search_replace.]"
+                        )
+                        # Also inject as system note — harder for model to ignore
+                        self._inject_system_note(
+                            f"STOP running bash. The error is at {tb_file}:{tb_line}. "
+                            f"Use read_file to read that file, then search_replace to fix the bug. "
+                            f"Do NOT run another bash command until you have fixed the code."
                         )
 
             yield ToolResultEvent(
@@ -1584,6 +1642,106 @@ class AgentLoop:
 
         return None
 
+    def _build_auto_context(self, user_msg: str) -> str | None:
+        """Build auto-delegation context based on the prompt and project state.
+
+        Instead of hoping the model calls task()/invoke_skill(), we inject:
+        1. Project file listing (if files exist) — replaces explore subagent
+        2. Skill content (if prompt matches a skill) — replaces invoke_skill
+        3. Planning prompt (if complex build task) — replaces planner subagent
+        """
+        parts: list[str] = []
+        msg_lower = user_msg.lower()
+
+        # 1. Auto-explore: list project files so model doesn't have to
+        try:
+            cwd = Path.cwd()
+            py_files = sorted(cwd.rglob("*.py"))
+            py_files = [f for f in py_files if ".logs" not in str(f) and ".venv" not in str(f)]
+            if len(py_files) >= 3:
+                listing = "\n".join(f"  {f.relative_to(cwd)}" for f in py_files[:30])
+                parts.append(f"PROJECT FILES ({len(py_files)} Python files):\n{listing}")
+                if len(py_files) > 30:
+                    parts.append(f"  ... and {len(py_files) - 30} more")
+        except Exception:
+            pass
+
+        # Also list non-Python files of interest
+        try:
+            for pattern in ("*.md", "*.txt", "*.json", "*.yaml", "*.yml", "*.toml", "*.csv"):
+                for f in sorted(cwd.glob(pattern))[:5]:
+                    if f.name not in (".logs",):
+                        parts.append(f"  {f.name} ({f.stat().st_size} bytes)")
+        except Exception:
+            pass
+
+        # 2. Skill routing: match prompt to skill and inject content
+        skill_name = None
+        if any(kw in msg_lower for kw in ("review this", "review the", "code review", "audit")):
+            skill_name = "review"
+        elif any(kw in msg_lower for kw in ("investigate", "debug", "why does", "why is", "broken", "diagnose")):
+            skill_name = "investigate"
+        elif any(kw in msg_lower for kw in ("ship it", "commit and push", "create a pr", "open a pr")):
+            skill_name = "ship"
+        elif any(kw in msg_lower for kw in ("simplify", "clean up", "reduce complexity")):
+            skill_name = "simplify"
+
+        if skill_name:
+            skill_info = self.skill_manager.get_skill(skill_name)
+            if skill_info and skill_info.skill_path:
+                try:
+                    content = skill_info.skill_path.read_text(encoding="utf-8")
+                    # Strip frontmatter
+                    if content.startswith("---"):
+                        end = content.find("---", 3)
+                        if end > 0:
+                            content = content[end + 3:].strip()
+                    parts.append(f"FOLLOW THIS WORKFLOW ({skill_name} skill):\n{content[:2000]}")
+                except Exception:
+                    pass
+
+        # 3. Planning nudge for complex build tasks
+        is_build = any(kw in msg_lower for kw in (
+            "build", "create", "implement", "make a", "write a",
+            "set up", "scaffold", "get started", "prd",
+        ))
+        is_complex = len(user_msg) > 100 or any(kw in msg_lower for kw in (
+            "multiple", "modules", "package", "api", "database",
+            "features", "cli", "commands",
+        ))
+        if is_build and is_complex:
+            parts.append(
+                "PLANNING REQUIRED — This is a complex build task. Before writing code:\n"
+                "1. List ALL files you will create (with brief purpose)\n"
+                "2. Define the package structure\n"
+                "3. Use ABSOLUTE imports (from pkg.module import X), NOT relative imports\n"
+                "4. Create files in dependency order (utilities first, then modules that import them)\n"
+                "5. After creating all files, test with: python3 -m package_name\n"
+                "6. If tests fail, READ the error, then FIX with search_replace"
+            )
+        elif is_build:
+            parts.append(
+                "REMINDER: Use write_file to create files. Use absolute imports. "
+                "Test with python3 -m package_name (not python3 package/file.py)."
+            )
+
+        if not parts:
+            return None
+        return "\n\n".join(parts)
+
+    def _recent_tool_names(self, limit: int = 10) -> list[str]:
+        """Return recent tool names from message history (most recent last)."""
+        names: list[str] = []
+        for msg in reversed(self.messages):
+            if msg.role == Role.assistant and msg.tool_calls:
+                for tc in msg.tool_calls:
+                    if tc.function and tc.function.name:
+                        names.append(tc.function.name)
+            if len(names) >= limit:
+                break
+        names.reverse()
+        return names
+
     def _prune_repeated_tool_calls(self) -> None:
         """Remove duplicate tool call/result pairs from recent history.
 

{drydock_cli-1.5.0 → drydock_cli-1.5.2}/drydock/core/prompts/cli.md

@@ -223,6 +223,13 @@ Error Handling:
 - After ANY error, your next action must be DIFFERENT from what just failed.
 - Never run the same failing command twice — the result won't change.
 
+Python Package Rules:
+When creating Python packages (directories with __init__.py):
+- Use ABSOLUTE imports, not relative. Write `from mypackage.module import X`, NOT `from .module import X`
+- To run a package: `python3 -m mypackage` (requires __main__.py), NOT `python3 mypackage/main.py`
+- Relative imports (from .foo) ONLY work with `python3 -m`, not direct file execution
+- If you get "attempted relative import with no known parent package", switch to absolute imports
+
 Debugging Best Practices:
 When fixing bugs, follow this cycle: RUN → READ → FIX → VERIFY.
 1. RUN the failing command to see the exact error (traceback, assertion, etc.)
@@ -231,7 +238,7 @@ When fixing bugs, follow this cycle: RUN → READ → FIX → VERIFY.
    - TypeError (str + int): add type conversion (int(), float(), str())
    - KeyError: add .get() with default, or check key existence
    - IndexError: fix loop bounds or add bounds check
-   - ImportError: fix the import path or module name
+   - ImportError: fix the import path, use absolute imports, or run with python3 -m
    - Off-by-one: check if index starts at 0 or 1
    - Missing return value: check if function returns in all paths
 4. VERIFY by running the command again — if it still fails, read the NEW error

{drydock_cli-1.5.0 → drydock_cli-1.5.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "1.5.0"
+version = "1.5.2"
 description = "Drydock — nautical CLI coding agent"
 readme = "README.md"
 requires-python = ">=3.12"

drydock_cli-1.5.2/scripts/monitor_test_battery.sh

@@ -0,0 +1,115 @@
+#!/bin/bash
+# Monitor the overnight PRD test battery.
+# Restarts if stuck, logs results, and stops on repeated failures.
+# Cron: */10 * * * * /data3/drydock/scripts/monitor_test_battery.sh
+
+set -euo pipefail
+
+export PATH="/home/bobef/miniconda3/bin:$PATH"
+PYTHON="/home/bobef/miniconda3/bin/python3"
+DRYDOCK="/data3/drydock"
+LOGDIR="$DRYDOCK/test_bank_results"
+LOGFILE="$LOGDIR/overnight_run2.log"
+MONITOR_LOG="$LOGDIR/monitor.log"
+LOCKFILE="/tmp/drydock_test_monitor.lock"
+
+log() { echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*" >> "$MONITOR_LOG"; }
+
+# Prevent concurrent monitors
+if [ -f "$LOCKFILE" ]; then
+    pid=$(cat "$LOCKFILE" 2>/dev/null)
+    if kill -0 "$pid" 2>/dev/null; then
+        exit 0
+    fi
+fi
+echo $$ > "$LOCKFILE"
+trap "rm -f $LOCKFILE" EXIT
+
+cd "$DRYDOCK"
+
+# Check if test process is running
+TEST_PID=$(pgrep -f "pytest.*test_bank_prd" 2>/dev/null | head -1 || echo "")
+
+if [ -z "$TEST_PID" ]; then
+    # Check if it completed
+    if [ -f "$LOGFILE" ] && grep -q "passed" "$LOGFILE" 2>/dev/null; then
+        PASSED=$(grep -c "PASSED" "$LOGFILE" 2>/dev/null || echo 0)
+        FAILED=$(grep -c "FAILED" "$LOGFILE" 2>/dev/null || echo 0)
+        TOTAL=$((PASSED + FAILED))
+
+        if [ "$TOTAL" -ge 65 ]; then
+            log "TEST BATTERY COMPLETE: $PASSED passed, $FAILED failed out of $TOTAL"
+            # Log failures
+            grep "FAILED" "$LOGFILE" >> "$MONITOR_LOG" 2>/dev/null || true
+            exit 0
+        fi
+    fi
+
+    # Not running and not complete — restart
+    log "Test battery not running. Restarting..."
+
+    # Check how many tests already completed
+    DONE=$(grep -c "PASSED\|FAILED" "$LOGFILE" 2>/dev/null || echo 0)
+
+    if [ "$DONE" -gt 0 ]; then
+        # Save progress
+        cp "$LOGFILE" "$LOGDIR/overnight_run2_partial_$(date +%H%M).log" 2>/dev/null || true
+
+        # Check for repeated failures (same test failing 3+ times across runs)
+        FAIL_LOG="$LOGDIR/failure_tracking.log"
+        grep "FAILED" "$LOGFILE" >> "$FAIL_LOG" 2>/dev/null || true
+
+        # If a specific test has failed 3+ times, skip it
+        SKIP_TESTS=""
+        if [ -f "$FAIL_LOG" ]; then
+            while IFS= read -r test_name; do
+                COUNT=$(grep -c "$test_name" "$FAIL_LOG" 2>/dev/null || echo 0)
+                if [ "$COUNT" -ge 3 ]; then
+                    log "SKIP: $test_name failed $COUNT times, skipping"
+                    SKIP_TESTS="$SKIP_TESTS and not ${test_name##*::}"
+                fi
+            done < <(grep "FAILED" "$FAIL_LOG" | sed 's/ .*//' | sort -u)
+        fi
+    fi
+
+    # Restart
+    DESELECT=""
+    if [ -n "${SKIP_TESTS:-}" ]; then
+        DESELECT="-k 'not _PLACEHOLDER_ $SKIP_TESTS'"
+    fi
+
+    nohup $PYTHON -m pytest tests/test_bank_prd.py tests/test_bank_prd_extended.py \
+        -v -p no:xdist -p no:cov --override-ini="addopts=" --tb=short \
+        $DESELECT \
+        > "$LOGFILE" 2>&1 &
+
+    log "Restarted with PID $! (skipping: ${SKIP_TESTS:-none})"
+    exit 0
+fi
+
+# Process is running — check if stuck
+LAST_MODIFIED=$(stat -c %Y "$LOGFILE" 2>/dev/null || echo 0)
+NOW=$(date +%s)
+STALE=$((NOW - LAST_MODIFIED))
+
+if [ "$STALE" -gt 600 ]; then
+    # Log file hasn't been updated in 10 minutes — test might be stuck
+    CURRENT_TEST=$(tail -1 "$LOGFILE" | grep -o "test_[a-z_]*" | head -1 || echo "unknown")
+    log "WARNING: Test battery stale for ${STALE}s. Current test: $CURRENT_TEST"
+
+    if [ "$STALE" -gt 1200 ]; then
+        # 20 minutes stuck — kill and restart
+        log "STUCK: Killing PID $TEST_PID (stale ${STALE}s on $CURRENT_TEST)"
+        kill "$TEST_PID" 2>/dev/null || true
+        sleep 5
+        kill -9 "$TEST_PID" 2>/dev/null || true
+
+        # Will restart on next cron run
+    fi
+else
+    # Running fine — log progress
+    PASSED=$(grep -c "PASSED" "$LOGFILE" 2>/dev/null || echo 0)
+    FAILED=$(grep -c "FAILED" "$LOGFILE" 2>/dev/null || echo 0)
+    TOTAL=$((PASSED + FAILED))
+    log "Running: $TOTAL/65 done ($PASSED passed, $FAILED failed)"
+fi