Reference implementation for Agent Skills v1.1 — validates skills, checks dependencies, and runs tests.
pip install -r requirements.txt# Check if skill is valid (dependencies present, versions OK)
python skills_ref.py validate ./my-skill --skills-root ./skills
# Force validation even with missing deps (for development)
python skills_ref.py validate ./my-skill --skills-root ./skills --force# Creates SKILL.md template with auto-populated requires
python skills_ref.py init ./my-new-skill --skills-root ./skillspython skills_ref.py test ./my-skill --skills-root ./skills# List all skills
python skills_ref.py deps --skills-root ./skills
# Show dependency graph
python skills_ref.py deps --skills-root ./skills --graph
# Check for circular dependencies
python skills_ref.py deps --skills-root ./skills --check-circularTest cases are defined in YAML:
cases:
- name: test_name
description: What this test verifies
input: "The prompt to send to the agent"
assertions:
output_contains:
- "expected text"
output_not_contains:
- "error"
output_matches:
- "regex pattern"
semantic_match:
criterion: "Clear criterion for LLM judge"To actually run tests against an agent, pass an agent_runner function to run_tests():
from skills_ref import run_tests
def my_agent_runner(skill_path, input_prompt):
# Call your agent here
# Return the output as a string
return agent.run(skill_path, input_prompt)
passed, total = run_tests(skill_path, skills_root, agent_runner=my_agent_runner)This project proposes extensions to the Agent Skills specification for dependency management and testing.
See the full RFC: RFC-001-dependencies-testing.md
Related discussions:
Apache 2.0