Thank you for your interest in contributing to Py-Agent! We welcome contributions of all kinds: bug fixes, new features, new agent skills/profiles, documentation improvements, and local model benchmark reports.
When submitting code or features, please keep our core philosophy in mind:
- Lightweight: 0% idle CPU/RAM when not actively running. Standard Python library execution where possible.
- Local-First & Private: Local inference (
llama-server) and offline fallbacks are always first-class citizens. - Zero-Trust Safety: Never bypass path-boundary checks or execution confirmation gates.
- Token Efficiency: Prompts, context loaders, and tool outputs must be concise and avoid unnecessary context bloat.
git clone https://github.com/YOUR_USERNAME/local-ai.git
cd local-aiOn Arch Linux:
sudo pacman -S python-rich python-requests
yay -S python-sqlite-vec python-textual python-uvloopBefore making any changes, verify that your environment passes all health checks:
./tools/test-agentYou can contribute new specialized agent skills:
- Place skills in
skills/<category>/<skill_name>.md. - Follow the YAML frontmatter standard with clear system instructions and role constraints.
Tools in tools/agentic/ or tools/subsec/ should:
- Be standalone, executable scripts.
- Handle missing dependencies gracefully with helpful error messages.
Share prompt optimizations or setup/*.sh launch scripts for newly released quantized GGUF models.
- Branch Naming: Use clear branch names:
feat/your-feature-namefix/bug-descriptiondocs/improvement
- Run Tests: Ensure
./tools/test-agentpasses with no broken paths or failed imports. - Commit Messages: Follow conventional commit style:
feat: add new NOOA kernel preview methodfix: prevent race condition in tpm memory compilerdocs: update roadmap for v1.0.0
- Open a PR: Describe the changes clearly and link any relevant issues.
By contributing to Py-Agent, you agree that your contributions will be licensed under the project's MODIFIED MIT LICENSE.