Thank you for your interest in improving AI agent skills for better code quality!
This repository is organized around self-documenting skills that include automated benchmark tests to ensure reliability across different AI models.
basic-programming-skills/
├── skills/ # All skill definitions
│ ├── ps-skill-name/
│ │ ├── SKILL.md # Documentation & Examples
│ │ └── test.json # Benchmark scenarios
├── tests/ # Benchmarking engine
└── docs/ # Documentation
cd skills
mkdir ps-your-skill-name # Use lowercase-with-hyphens
cd ps-your-skill-nameEvery skill needs a SKILL.md with YAML frontmatter:
---
name: ps-your-skill-name
description: Short description for AI context discovery.
severity: BLOCK | WARN | SUGGEST
---
# Your Skill Name
## Principle
What core rule does this skill enforce?
## When to Use
List specific scenarios or code smells.
## Instructions
✅ **Do:** Clear guidelines.
❌ **Avoid:** Common anti-patterns.
## Examples (Pseudocode)
Use language-agnostic pseudocode for principles.Benchmark scenarios ensure the skill works. Tests use the following schema:
[
{
"name": "scenario_name",
"input": "Code snippet or external_file.js",
"expected": {
"includes": ["term_must_exist"],
"excludes": ["forbidden_pattern"],
"regex": ["pattern\\d+"],
"min_length": 50
}
}
]Before submitting, verify your skill with the built-in evaluator:
# Run benchmark for your specific skill
uv run python tests/evaluator.py --skill ps-your-skill-name --verbose- Correct
severityassigned inSKILL.md. - At least 2 test scenarios in
test.json. - Pass rate > 50% for the "With Skill" run.
- No grammar or spelling errors in documentation.
- Fork the repository.
- Add your skill folder.
- Update
README.mdandCHANGELOG.md. - Submit PR with a clear description of the new principle.
By contributing, you agree your contributions will be licensed under the MIT License.