The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from files and agent messages to generate readability reports. \n
Ingestion points: The script hooks/readability_hook.py ingests the assistant's last message and contents of files modified by the agent's write or edit tools. The CLI tool resources/readability.py reads from files or standard input.\n
Boundary markers: Findings are returned to the agent in a structured format with clear headers (e.g., "Readability check failed") defining the report's purpose.\n
Capability inventory: The skill is restricted to reading local files, performing text calculations, and writing a non-sensitive status marker file to the system temporary directory for loop protection. It performs no network operations.\n
Sanitization: The hook script implements sanitization by truncating failure evidence (quoted sentences) to a maximum of 200 characters, mitigating the risk of injecting long, instruction-carrying text back into the prompt context.\n- [DYNAMIC_EXECUTION]: The hook script uses Python's importlib to dynamically load the readability.py scoring engine from a relative path within the skill's own directory. This is a standard method for modular code organization and does not involve executing external or user-provided code.

readability-check