The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/validate.py script executes several subprocesses for validation tasks:
It invokes ruby -e with a hardcoded script to perform secure YAML parsing of skill frontmatter using YAML.safe_load.
It uses python3 -m py_compile to check for syntax errors in Python scripts within the audited skill.
It uses bash -n to perform linting and syntax checks on shell scripts.
[COMMAND_EXECUTION]: The references/claude.md adapter provides instructions and a command template for creating symlinks (ln -s) to manage skill discovery paths on the local filesystem.
[PROMPT_INJECTION]: The skill instructions (Category 8: Indirect Prompt Injection) involve processing untrusted third-party skill files.
Ingestion points: Reads SKILL.md and auxiliary resource files from a target skill directory provided by the user.
Boundary markers: The skill does not explicitly define delimiters for audited content, but the processing is structured via the validate.py tool rather than raw inclusion in the agent's context.
Capability inventory: File system read access, subprocess execution for linting, and symlink creation.
Sanitization: The validator specifically uses secure YAML loading in Ruby to prevent exploitation of the parser via malicious frontmatter.
[SAFE]: The skill includes a defensive SECRET_RE regex pattern in scripts/validate.py designed to detect and alert users about hardcoded credentials or API keys in the skills being audited.

x9-skill-creator