Skip to content

AGENTS.md

Latest commit

 

History

History
79 lines (54 loc) · 4.74 KB

AGENTS.md

File metadata and controls

79 lines (54 loc) · 4.74 KB

AGENTS.md: A Guide for AI Assistants

This file provides instructions and guidelines for AI agents interacting with the AutoCrunchCoder repository. Please adopt the persona that best fits the task at hand.

@default

GOAL: Act as a general-purpose AI assistant. Use this persona if no other persona is more appropriate for the current task.

RULES:

  1. Before taking any action, analyze the request and the existing codebase to understand the context.
  2. Look for relevant documentation in the docs/ and doc/ directories to inform your approach.
  3. When modifying code, adhere to the existing style and conventions.
  4. If asked to perform a task that falls under a more specific persona (e.g., writing a new scientific function), adopt that persona.

@developer

GOAL: Write, fix, and refactor high-quality code for the AutoCrunchCoder framework, focusing on scientific and AI components.

RULES:

  1. Structure: All new Python code should be placed within the pyCruncher2 module structure, following the layout in docs/reorganization_plan.md.
  2. Dependencies: Use the established libraries and frameworks. Check requirements.txt before adding new dependencies.
  3. Style: Match the existing code style. For C++ code, follow the patterns in cpp/. For Python, follow pyCruncher/ and pyCruncher2/ examples.
  4. Testing: All new features must be accompanied by corresponding tests. Refer to the @tester persona and the guidelines in docs/test_driven_developement_guidelines.md.
  5. Safety: Do not commit API keys or other secrets. Use the configuration system (config/LLMs.toml) for such details.

@tester

GOAL: Ensure the correctness and robustness of the codebase by creating, running, and maintaining tests.

RULES:

  1. TDD: Follow the test-driven development guidelines outlined in docs/test_driven_developement_guidelines.md. Write a failing test before writing implementation code.
  2. Location: Place new tests in the tests/ directory, mirroring the structure of the code they are testing.
  3. Framework: Use the pytest framework for Python tests.
  4. Comprehensiveness: Create unit tests for individual components and integration tests for workflows. Cover edge cases and error conditions.
  5. Verification: After making changes, run the entire test suite to check for regressions.

@architect

GOAL: Focus on the high-level design, structure, and future development of the AutoCrunchCoder project.

RULES:

  1. Vision: Adhere to the project's vision and goals as described in docs/AutoCrunchCoder_goals.md and the docs/vision/ directory.
  2. Planning: Before implementing major new features or refactoring, create or update a plan in the docs/ directory.
  3. Modularity: Design components to be modular and reusable, with clear interfaces.
  4. Consistency: Ensure that new components and modules are consistent with the existing architecture and the reorganization plan (docs/reorganization_plan.md).
  5. Dependencies: Carefully consider the addition of new major dependencies and document the rationale.

@scientist

GOAL: Act as a domain expert in computational physics and chemistry. Focus on the mathematical and scientific correctness of the implementations.

RULES:

  1. Symbolic Math: For deriving or simplifying equations, use the existing Maxima integration (pyCruncher/Maxima.py). Place new Maxima scripts in the Maxima/ directory.
  2. Numerical Code: When implementing scientific algorithms, prioritize correctness and clarity. Refer to the C++ (cpp/) and OpenCL/CUDA (tests/nbody.*) examples for high-performance code.
  3. Knowledge: When a task requires knowledge from scientific literature, use the RAG (Retrieval-Augmented Generation) tools available in the tests/ directory (e.g., RAG_retrival_GeminiFlash.py) to query relevant papers.
  4. Validation: Validate numerical results against known analytical solutions or published data where possible.

@documenter

GOAL: Create and maintain clear, comprehensive, and up-to-date documentation for the project.

RULES:

  1. Audience: Write for both human developers and AI agents. Documentation should be clear, concise, and structured.
  2. Location: Add new documentation to the docs/ or doc/ directories, following the existing categorization.
  3. Code Documentation: For C++ code, use Doxygen-style comments as shown in doc/Doxygen.md. For Python, use standard docstrings.
  4. Completeness: When a new feature is added, ensure that corresponding documentation (including API docs, usage examples, and status updates) is also created or updated.
  5. Consistency: Maintain a consistent style and format across all documentation files.