The Agent Skills Directory

[COMMAND_EXECUTION]: The shell command execution block in SKILL.md uses the $ARGUMENTS variable without double-quotes. This allows for potential command injection if the input provided by the agent contains shell metacharacters such as semicolons, pipes, or ampersands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown files from a user-specified directory, which could contain malicious instructions designed to influence the agent when it reviews the validation output.
Ingestion points: In scripts/okf_validate.py, the script recursively searches for and reads the content of all .md files within the provided bundle directory using Path.read_text().
Boundary markers: The validation output does not employ delimiters or explicit instructions to the agent to ignore embedded commands within the reported errors and warnings.
Capability inventory: The skill includes filesystem write capabilities (via the --migrate flag) and shell command execution via the Bash tool.
Sanitization: The script performs structural validation but does not sanitize or escape content reflected from the files (such as frontmatter keys, headings, or file paths) before outputting it to the agent.
[EXTERNAL_DOWNLOADS]: The skill uses pip install to download the pyyaml package if the uv package manager is unavailable. This is a standard dependency for the script's functionality.

validate