agentic-python-coder 2.2.0__tar.gz → 2.3.0__tar.gz

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/.gitignore

@@ -152,6 +152,7 @@ cython_debug/
 uv.lock
 
 # Project specific
+coder-examples/
 CLAUDE-archive.md
 PROCESS_NOTES.md
 conversation_log.json

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentic-python-coder
-Version: 2.2.0
+Version: 2.3.0
 Summary: A lightweight Python coding agent that writes, executes, and iterates on code through natural language instructions
 Author: Stefan Szeider
 License: Apache-2.0
@@ -14,25 +14,26 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Topic :: Software Development :: Code Generators
 Requires-Python: <3.14,>=3.13
-Requires-Dist: ipykernel>=6.30.1
-Requires-Dist: jupyter-client>=8.6.3
-Requires-Dist: langchain-anthropic>=1.2.0
-Requires-Dist: langchain-core>=1.1.0
-Requires-Dist: langchain-experimental>=0.4.0
-Requires-Dist: langchain-openai>=1.1.0
-Requires-Dist: langgraph>=1.0.4
+Requires-Dist: ipykernel>=7.0.0
+Requires-Dist: jupyter-client>=8.7.0
+Requires-Dist: langchain-anthropic>=1.3.0
+Requires-Dist: langchain-core>=1.2.0
+Requires-Dist: langchain-experimental>=0.4.1
+Requires-Dist: langchain-openai>=1.1.6
+Requires-Dist: langchain>=1.2.0
+Requires-Dist: langgraph>=1.0.5
 Requires-Dist: mcp>=1.0.0
 Requires-Dist: python-dotenv>=1.2.1
 Requires-Dist: pyyaml>=6.0.3
 Requires-Dist: rich>=14.2.0
 Provides-Extra: dev
-Requires-Dist: mypy>=1.19.0; extra == 'dev'
-Requires-Dist: ruff>=0.14.7; extra == 'dev'
+Requires-Dist: mypy>=1.19.1; extra == 'dev'
+Requires-Dist: ruff>=0.14.10; extra == 'dev'
 Provides-Extra: test
-Requires-Dist: pytest-asyncio>=1.2.0; extra == 'test'
+Requires-Dist: pytest-asyncio>=1.3.0; extra == 'test'
 Requires-Dist: pytest-cov>=7.0.0; extra == 'test'
 Requires-Dist: pytest-watch>=4.2.0; extra == 'test'
-Requires-Dist: pytest>=9.0.1; extra == 'test'
+Requires-Dist: pytest>=9.0.2; extra == 'test'
 Description-Content-Type: text/markdown
 
 # Agentic Python Coder
@@ -291,17 +292,32 @@ Add to your MCP settings (e.g., `~/.claude/claude_desktop_config.json` or projec
 | Tool | Description |
 |------|-------------|
 | `python_exec` | Execute Python code. Auto-starts session if needed. Default 30s timeout. |
-| `python_reset` | Clear session state. Optionally install packages (e.g., `packages=["numpy", "pandas"]`). |
-| `python_status` | Check if session is active, Python version, installed packages, defined variables. |
+| `python_reset` | Create new kernel (no `kernel_id`) OR reset existing kernel (with `kernel_id`). Optionally install packages. |
+| `python_status` | Check session state: active flag, all active kernel IDs, Python version, packages, variables. |
 | `python_interrupt` | Send interrupt signal to stop long-running code. Session state is preserved. |
 
+### Multi-Agent Workflow
+
+For parallel agents, each agent gets its own kernel:
+
+```
+Agent A                              Agent B
+────────                             ────────
+python_reset() → kernel_id="aaa"     python_reset() → kernel_id="bbb"
+python_exec(kernel_id="aaa", ...)    python_exec(kernel_id="bbb", ...)
+python_exec(kernel_id="aaa", ...)    python_exec(kernel_id="bbb", ...)
+```
+
+Simple single-agent use: just call `python_exec()` — the default kernel auto-starts.
+
 ### Features
 
 - **Persistent state**: Variables, imports, and definitions persist across executions
-- **Auto-start**: Session starts automatically on first `python_exec`
+- **Auto-start**: Default session starts automatically on first `python_exec`
 - **Package installation**: Use `python_reset` with `packages` parameter to install dependencies
 - **Timeout handling**: Long-running code times out gracefully (session preserved)
 - **Interrupt support**: Stop runaway code without losing session state
+- **Multi-kernel**: Each `python_reset()` creates an isolated kernel for parallel agents
 
 ### Usage Tips
 

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/README.md

@@ -254,17 +254,32 @@ Add to your MCP settings (e.g., `~/.claude/claude_desktop_config.json` or projec
 | Tool | Description |
 |------|-------------|
 | `python_exec` | Execute Python code. Auto-starts session if needed. Default 30s timeout. |
-| `python_reset` | Clear session state. Optionally install packages (e.g., `packages=["numpy", "pandas"]`). |
-| `python_status` | Check if session is active, Python version, installed packages, defined variables. |
+| `python_reset` | Create new kernel (no `kernel_id`) OR reset existing kernel (with `kernel_id`). Optionally install packages. |
+| `python_status` | Check session state: active flag, all active kernel IDs, Python version, packages, variables. |
 | `python_interrupt` | Send interrupt signal to stop long-running code. Session state is preserved. |
 
+### Multi-Agent Workflow
+
+For parallel agents, each agent gets its own kernel:
+
+```
+Agent A                              Agent B
+────────                             ────────
+python_reset() → kernel_id="aaa"     python_reset() → kernel_id="bbb"
+python_exec(kernel_id="aaa", ...)    python_exec(kernel_id="bbb", ...)
+python_exec(kernel_id="aaa", ...)    python_exec(kernel_id="bbb", ...)
+```
+
+Simple single-agent use: just call `python_exec()` — the default kernel auto-starts.
+
 ### Features
 
 - **Persistent state**: Variables, imports, and definitions persist across executions
-- **Auto-start**: Session starts automatically on first `python_exec`
+- **Auto-start**: Default session starts automatically on first `python_exec`
 - **Package installation**: Use `python_reset` with `packages` parameter to install dependencies
 - **Timeout handling**: Long-running code times out gracefully (session preserved)
 - **Interrupt support**: Stop runaway code without losing session state
+- **Multi-kernel**: Each `python_reset()` creates an isolated kernel for parallel agents
 
 ### Usage Tips
 

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/coder/prompts/system.md

@@ -16,19 +16,12 @@ You have access to these specialized tools:
    - Call this ONCE when you have a complete, working solution
    - The code will be saved to {basename}_code.py
 
-3. **report_issue**: Provide feedback and summary
-   - Use at the end to summarize what was accomplished
-   - Report any issues, ambiguities, or difficulties encountered
-   - Also report if everything worked perfectly
-   - This feedback will be included in the log file
-
 ## Workflow
 
 1. **Understand the Task**: Read the problem in the <task> section carefully
 2. **Plan Your Approach**: Think through the problem and plan your solution strategy
 3. **Develop Solution**: Use python_exec iteratively to build and test
 4. **Save Final Code**: Call save_code with your complete solution
-5. **Provide Feedback**: Use report_issue to summarize and provide feedback
 
 ## Python Execution Best Practices
 
@@ -68,8 +61,7 @@ Build solutions incrementally:
 1. **Focus on the Task**: Complete what's requested, nothing more
 2. **Test Efficiently**: One or two test cases are usually sufficient
 3. **Save Once**: Call save_code only when you have the final code
-4. **Always Provide Feedback**: Use report_issue at the end to summarize your work
-5. **Stop When Done**: Don't add features not requested
+4. **Stop When Done**: Don't add features not requested
 
 ## Error Recovery
 
@@ -94,10 +86,6 @@ When finishing:
 1. Verify the solution works correctly
 2. Clean the code according to the **Code Cleaning Requirements** above
 3. Call save_code with the complete, cleaned code
-4. Call report_issue to provide your final summary and feedback:
-   - Summarize what was accomplished
-   - Report any issues, ambiguities, or difficulties encountered
-   - Even if everything worked perfectly, report: "All is fine - no issues encountered."
-5. STOP - do not continue unless asked
+4. STOP - do not continue unless asked
 
-Your goal is efficient, focused problem-solving.
+Your goal is efficient, focused problem-solving.

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/coder/prompts/system_todo.md

@@ -16,35 +16,28 @@ You have access to these specialized tools:
    - Call this ONCE when you have a complete, working solution
    - The code will be saved to {basename}_code.py
 
-3. **report_issue**: Provide feedback and summary
-   - Use at the end to summarize what was accomplished
-   - Report any issues, ambiguities, or difficulties encountered
-   - Also report if everything worked perfectly
-   - This feedback will be included in the log file
-
-4. **todo_write**: MANDATORY task management tool
+3. **todo_write**: MANDATORY task management tool
    - You MUST use this tool after understanding the problem
    - Create a todo list with items appropriate to the problem complexity
      (ranging from 3 simple items to over a dozen for complex problems)
    - Update todo item status as you progress (pending → in_progress → completed)
    - Only ONE todo item can be in_progress at a time
    - Add new todo items if you discover additional work needed
-   - Your last three todo items should always be: "Clean final code", "Save final code", and "Provide feedback"
+   - Your last two todo items should always be: "Clean final code" and "Save final code"
 
 ## Workflow
 
 1. **Understand the Task**: Read the problem in the <task> section carefully
-2. **Plan Your Approach** (MANDATORY): 
+2. **Plan Your Approach** (MANDATORY):
    - Use todo_write to create your todo list based on your understanding
    - The number of todo items should match problem complexity (3-12+ items)
-   - Include "Save final code" and "Provide feedback" as your last two items
+   - Include "Save final code" as your last item
    - This demonstrates planning and helps track progress
 3. **Develop Solution**: Use python_exec iteratively to build and test
    - Mark todo items as in_progress when starting them
    - Mark as completed when done
-4. **Clean Final Code**: Clean the code according to Code Cleaning Requirements (third-to-last todo item)
-5. **Save Final Code**: Call save_code with your complete, cleaned solution (second-to-last todo item)
-6. **Provide Feedback**: Use report_issue to summarize and provide feedback (final todo item)
+4. **Clean Final Code**: Clean the code according to Code Cleaning Requirements (second-to-last todo item)
+5. **Save Final Code**: Call save_code with your complete, cleaned solution (final todo item)
 
 ## Python Execution Best Practices
 
@@ -85,8 +78,7 @@ Build solutions incrementally:
 2. **Focus on the Task**: Complete what's requested, nothing more
 3. **Test Efficiently**: One or two test cases are usually sufficient
 4. **Save Once**: Call save_code only when you have the final code
-5. **Always Provide Feedback**: Use report_issue at the end to summarize your work
-6. **Stop When Done**: Don't add features not requested
+5. **Stop When Done**: Don't add features not requested
 
 ## Error Recovery
 
@@ -111,13 +103,9 @@ When finishing (these should be your final todo items):
 1. Ensure all todo items are marked as completed
 2. Verify the solution works correctly
 3. Clean the code according to the **Code Cleaning Requirements** above
-4. Call save_code with the complete, cleaned code (second-to-last todo item)
-5. Call report_issue to provide your final summary and feedback (final todo item):
-   - Summarize what was accomplished
-   - Report any issues, ambiguities, or difficulties encountered
-   - Even if everything worked perfectly, report: "All is fine - no issues encountered."
-6. STOP - do not continue unless asked
+4. Call save_code with the complete, cleaned code (final todo item)
+5. STOP - do not continue unless asked
 
 Note: Your todo list should show a clear progression from planning through completion.
 
-Your goal is efficient, focused problem-solving.
+Your goal is efficient, focused problem-solving.

agentic_python_coder-2.3.0/coder/src/agentic_python_coder/__init__.py

@@ -0,0 +1,74 @@
+"""Python Coding Agent - A minimal coding assistant using LangGraph and OpenRouter."""
+
+__version__ = "2.3.0"
+
+# High-level API (recommended for most users)
+from agentic_python_coder.runner import solve_task
+
+# Lower-level API (for custom workflows)
+from agentic_python_coder.agent import (
+    create_coding_agent,
+    run_agent,
+    get_final_response,
+    DEFAULT_STEP_LIMIT,
+)
+
+# LLM utilities
+from agentic_python_coder.llm import (
+    get_openrouter_llm,
+    load_model_config,
+    list_available_models,
+    DEFAULT_MODEL,
+)
+
+# Kernel management (multi-kernel API)
+from agentic_python_coder.kernel import (
+    # Core functions
+    create_kernel,
+    execute_in_kernel,
+    shutdown_kernel_by_id,
+    interrupt_kernel_by_id,
+    restart_kernel,
+    # Query functions
+    list_kernels,
+    kernel_exists,
+    get_kernel_info,
+    shutdown_all_kernels,
+    # Backward compat
+    get_kernel,
+    shutdown_kernel,
+    # Constants
+    DEFAULT_KERNEL_ID,
+    MAX_KERNELS,
+)
+
+__all__ = [
+    # Version
+    "__version__",
+    # High-level
+    "solve_task",
+    # Low-level agent
+    "create_coding_agent",
+    "run_agent",
+    "get_final_response",
+    "DEFAULT_STEP_LIMIT",
+    # LLM
+    "get_openrouter_llm",
+    "load_model_config",
+    "list_available_models",
+    "DEFAULT_MODEL",
+    # Kernel management
+    "create_kernel",
+    "execute_in_kernel",
+    "shutdown_kernel_by_id",
+    "interrupt_kernel_by_id",
+    "restart_kernel",
+    "list_kernels",
+    "kernel_exists",
+    "get_kernel_info",
+    "shutdown_all_kernels",
+    "get_kernel",
+    "shutdown_kernel",
+    "DEFAULT_KERNEL_ID",
+    "MAX_KERNELS",
+]

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/coder/src/agentic_python_coder/agent.py

@@ -1,8 +1,11 @@
 """ReAct agent for Python coding tasks."""
 
+import json
+import os
+import time
 from typing import Dict, Any, List, Optional
 from pathlib import Path
-from langgraph.prebuilt import create_react_agent
+from langchain.agents import create_agent
 from langgraph.checkpoint.memory import InMemorySaver
 
 from agentic_python_coder.llm import get_openrouter_llm
@@ -10,7 +13,6 @@ from agentic_python_coder.tools import (
     todo_write,
     python_exec,
     save_code,
-    report_issue,
     working_dir,
     set_task_basename,
     reset_global_state,
@@ -46,7 +48,7 @@ def create_coding_agent(
         working_directory: Directory for file operations
         system_prompt: System prompt as string (takes precedence over path)
         system_prompt_path: Path to system prompt file (used if system_prompt not provided)
-        model: Optional model name (defaults to claude-sonnet-4.5)
+        model: Optional model name (uses configured default if not specified)
         project_prompt: Optional project-specific prompt
         with_packages: Optional list of packages for dynamic mode
         task_content: Task description/content
@@ -67,8 +69,6 @@ def create_coding_agent(
 
     # Store packages for kernel initialization
     if with_packages is not None:
-        import os
-
         os.environ["CODER_WITH_PACKAGES"] = ",".join(with_packages)
 
     # Get LLM instance
@@ -80,9 +80,9 @@ def create_coding_agent(
 
     # Minimal tool set
     if todo:
-        tools = [python_exec, save_code, report_issue, todo_write]
+        tools = [python_exec, save_code, todo_write]
     else:
-        tools = [python_exec, save_code, report_issue]
+        tools = [python_exec, save_code]
 
     # Build combined prompt
     prompts = []
@@ -109,15 +109,13 @@ def create_coding_agent(
 
     # Create the agent with memory
     checkpointer = InMemorySaver()
-    agent = create_react_agent(
-        llm, tools, prompt=combined_prompt, checkpointer=checkpointer
+    agent = create_agent(
+        llm, tools, system_prompt=combined_prompt, checkpointer=checkpointer
     )
 
     # Store metadata for run_agent to use
     agent._coder_metadata = {
         "working_directory": working_directory,
-        "with_packages": with_packages,
-        "task_basename": task_basename,
     }
 
     return agent
@@ -127,24 +125,21 @@ def _print_tool_progress(tool_name: str, args: dict):
     """Print progress info for a tool call."""
     if tool_name == "python_exec" and "code" in args:
         code = args["code"]
+        code_stripped = code.strip()
         if "def " in code:
-            func_match = code.split("def ")[1].split("(")[0] if "def " in code else ""
+            func_match = code.split("def ")[1].split("(")[0]
             print(f"  {tool_name}: defining function {func_match}()")
         elif "class " in code:
-            class_match = (
-                code.split("class ")[1].split("(")[0].split(":")[0]
-                if "class " in code
-                else ""
-            )
+            class_match = code.split("class ")[1].split("(")[0].split(":")[0]
             print(f"  {tool_name}: defining class {class_match}")
-        elif "import " in code and len(code.strip().split("\n")) == 1:
-            print(f"  {tool_name}: {code.strip()}")
-        elif "=" in code and len(code.strip().split("\n")) == 1:
+        elif "import " in code and len(code_stripped.split("\n")) == 1:
+            print(f"  {tool_name}: {code_stripped}")
+        elif "=" in code and len(code_stripped.split("\n")) == 1:
             var_name = code.split("=")[0].strip()
             print(f"  {tool_name}: assigning variable {var_name}")
-        elif code.strip().startswith("print("):
+        elif code_stripped.startswith("print("):
             print(
-                f"  {tool_name}: {code.strip()[:50]}{'...' if len(code.strip()) > 50 else ''}"
+                f"  {tool_name}: {code_stripped[:50]}{'...' if len(code_stripped) > 50 else ''}"
             )
         elif "read_csv" in code or "read_excel" in code or "read_json" in code:
             print(f"  {tool_name}: loading data file")
@@ -171,32 +166,23 @@ def _print_tool_progress(tool_name: str, args: dict):
         print(f"\n  {tool_name}:")
         todos = args["todos"]
         for todo in todos:
-            status_symbol = (
-                "☒"
-                if todo["status"] == "completed"
-                else "☐"
-                if todo["status"] == "pending"
-                else "▶"
-            )
-            print(f"     {status_symbol} {todo['content']}")
+            status = todo.get("status", "")
+            content = todo.get("content", "")
+            status_symbol = "☒" if status == "completed" else "☐" if status == "pending" else "▶"
+            print(f"     {status_symbol} {content}")
     else:
-        arg_str = str(args)[:30]
-        if len(str(args)) > 30:
-            arg_str += "..."
-        print(f"  {tool_name}: {arg_str}")
+        args_str = str(args)
+        arg_display = args_str[:30] + "..." if len(args_str) > 30 else args_str
+        print(f"  {tool_name}: {arg_display}")
 
 
-def _process_tool_calls(msg, current_tools: dict, stats: dict, quiet: bool):
+def _process_tool_calls(msg, stats: dict, quiet: bool):
     """Process tool calls from a message, updating stats and optionally printing."""
     if hasattr(msg, "tool_calls") and msg.tool_calls:
         for tool_call in msg.tool_calls:
             tool_name = tool_call.get("name") or tool_call.get("function", {}).get(
                 "name"
             )
-            tool_id = tool_call.get("id")
-            if tool_id:
-                current_tools[tool_id] = tool_name
-
             if tool_name:
                 stats["tool_usage"][tool_name] = (
                     stats["tool_usage"].get(tool_name, 0) + 1
@@ -214,9 +200,6 @@ def _process_tool_calls(msg, current_tools: dict, stats: dict, quiet: bool):
         for tool_call in tool_calls:
             function = tool_call.get("function", {})
             tool_name = function.get("name")
-            tool_id = tool_call.get("id")
-            if tool_id and tool_name:
-                current_tools[tool_id] = tool_name
 
             if tool_name:
                 stats["tool_usage"][tool_name] = (
@@ -226,11 +209,9 @@ def _process_tool_calls(msg, current_tools: dict, stats: dict, quiet: bool):
             if not quiet and tool_name:
                 args_str = function.get("arguments", "{}")
                 try:
-                    import json
-
                     args = json.loads(args_str)
                     _print_tool_progress(tool_name, args)
-                except Exception:
+                except json.JSONDecodeError:
                     print(f"  {tool_name}")
 
 
@@ -244,6 +225,7 @@ def _update_token_stats(msg, stats: dict):
                 "completion_tokens", 0
             )
             stats["token_consumption"]["total_tokens"] += usage.get("total_tokens", 0)
+            return  # Avoid double-counting if both metadata sources exist
 
     if hasattr(msg, "usage_metadata") and msg.usage_metadata:
         stats["token_consumption"]["input_tokens"] += msg.usage_metadata.get(
@@ -286,7 +268,6 @@ def run_agent(
     config = {"configurable": {"thread_id": thread_id}, "recursion_limit": limit}
 
     messages = []
-    current_tools = {}
 
     # Initialize statistics
     stats = {
@@ -295,8 +276,6 @@ def run_agent(
         "execution_time_seconds": 0,
     }
 
-    import time
-
     start_time = time.time()
 
     # Stream the agent's work
@@ -314,7 +293,7 @@ def run_agent(
                     messages.append(msg)
 
                     # Process tool calls (always update stats, optionally print)
-                    _process_tool_calls(msg, current_tools, stats, quiet)
+                    _process_tool_calls(msg, stats, quiet)
 
                     # Update token statistics
                     _update_token_stats(msg, stats)

{agentic_python_coder-2.2.0 → agentic_python_coder-2.3.0}/coder/src/agentic_python_coder/cli.py

@@ -23,7 +23,11 @@ from agentic_python_coder.project_md import (
     check_packages_available,
     create_project_prompt,
 )
-from agentic_python_coder.llm import DEFAULT_MODEL, list_available_models, load_model_config
+from agentic_python_coder.llm import (
+    DEFAULT_MODEL,
+    list_available_models,
+    load_model_config,
+)
 from agentic_python_coder import __version__
 
 
@@ -88,7 +92,8 @@ def parse_args():
     )
 
     parser.add_argument(
-        "--version", "-V",
+        "--version",
+        "-V",
         action="version",
         version=f"%(prog)s {__version__}",
     )
@@ -106,7 +111,9 @@ def parse_args():
         help="Path to task file (creates {basename}_code.py and {basename}.jsonl)",
     )
 
-    parser.add_argument("--model", help=f"Model name or JSON file (default: {DEFAULT_MODEL})")
+    parser.add_argument(
+        "--model", help=f"Model name or JSON file (default: {DEFAULT_MODEL})"
+    )
 
     parser.add_argument(
         "--interactive", "-i", action="store_true", help="Interactive mode"
@@ -145,7 +152,8 @@ def parse_args():
     )
 
     parser.add_argument(
-        "--quiet", "-q",
+        "--quiet",
+        "-q",
         action="store_true",
         help="Suppress console output during execution",
     )
@@ -176,13 +184,13 @@ def copy_resource_dir(source_path, dest_path: Path):
         dest_item = dest_path / item.name
         if item.is_file():
             # Skip __pycache__ and .pyc files
-            if item.name.endswith('.pyc') or item.name == '__pycache__':
+            if item.name.endswith(".pyc") or item.name == "__pycache__":
                 continue
             # Read and write file content
             content = item.read_bytes()
             dest_item.write_bytes(content)
         elif item.is_dir():
-            if item.name == '__pycache__':
+            if item.name == "__pycache__":
                 continue
             copy_resource_dir(item, dest_item)
 
@@ -216,7 +224,9 @@ def init_examples(template: str = "all"):
 
     print(f"\nExamples initialized in: {output_dir.absolute()}")
     print("\nUsage:")
-    print(f"  coder --with cpmpy --project {output_dir}/cpmpy/cpmpy.md --task {output_dir}/cpmpy/sample_tasks/n_queens.md")
+    print(
+        f"  coder --with cpmpy --project {output_dir}/cpmpy/cpmpy.md --task {output_dir}/cpmpy/sample_tasks/n_queens.md"
+    )
 
 
 def validate_packages(packages):