The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell scripts such as scripts/validate_skill.sh and scripts/smoke_test.sh to perform static analysis on skill packages. These tools allow the agent to verify the structure and integrity of developed skills within the local development environment.
[DYNAMIC_EXECUTION]: The validator in scripts/validate_skill.py implements a custom YAML loader that inherits from yaml.SafeLoader. This ensures that parsing skill metadata does not result in the execution of arbitrary Python objects. The package includes unit tests in scripts/test_validate_skill.py that specifically confirm the rejection of malicious YAML tags.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as its core purpose is transforming untrusted user sessions into instructions.
Ingestion points: Data enters via session-derived workflows as documented in references/workflow-extraction.md.
Boundary markers: The instructions mandate separating reusable methods from private context and using synthetic examples.
Capability inventory: The agent can perform file writes to create skills and execute local validation scripts.
Sanitization: Guidelines require content review and abstraction to prevent instruction leakage from the source material.
[EXTERNAL_DOWNLOADS]: The skill references the PyYAML library (version 6.0.3) for metadata parsing. It provides instructions for downloading this dependency from official registries using the uv tool.

skill-builder