Browser4Programer

An AI programmer initially works for Browser4.

Auto-Coder: Self-Evolving Programming Automation

auto_coder is a Python automation pipeline that performs a closed-loop cycle of:

1. Understanding – Parse and analyze a programming task description
2. Design – Create a solution blueprint (parameters, algorithm steps)
3. Programming – Generate executable Python code
4. Execution – Run the generated code in a sandboxed namespace
5. Diagnosis – Identify root causes when execution fails
6. Automatic Repair – Patch the code and iterate until the task succeeds

Quick Start

# Run a task with a test assertion
python -m auto_coder "Write a function factorial that computes the factorial of n" \
    --test "factorial(5)"

# Verbose output
python -m auto_coder -v "Write a function add that adds two numbers" \
    --test "add(2, 3)"

Markdown Processing

Process code blocks inside a Markdown document through an automated Extract → Normalize → Execute → Diagnose → Auto-fix loop:

python -m auto_coder --markdown examples/demo.md
python -m auto_coder --md examples/demo.md -v --max-iterations 3

Each fenced Python code block is extracted, cleaned up, executed in a sandbox, and — if execution fails — automatically diagnosed and repaired. The loop repeats until the block succeeds or the iteration limit is reached.

Running Tests

pip install pytest
python -m pytest tests/ -v

Project Structure

src/auto_coder/
├── __init__.py            # Package metadata
├── __main__.py            # CLI entry point
├── understanding.py       # Stage 1: task parsing
├── design.py              # Stage 2: solution design
├── programming.py         # Stage 3: code generation
├── execution.py           # Stage 4: code execution
├── diagnosis.py           # Stage 5: error analysis
├── repair.py              # Stage 6: automatic repair
├── pipeline.py            # Orchestrator (closed loop)
└── markdown_processor.py  # Markdown code-block processing loop