The Agent Skills Directory

[DATA_EXFILTRATION]: The truncate_novel_from_file method in scripts/novel_truncator.py permits reading any file from the host system by providing an unvalidated file_path. This functionality can be exploited to access sensitive system files, environment variables, private keys, or credentials (e.g., .env, ~/.ssh/id_rsa, /etc/passwd).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text content provided as input, which creates a vulnerability to indirect prompt injection if the text contains malicious instructions for the agent.
Ingestion points: The novel_text parameter in scripts/novel_truncator.py and the main text input defined in SKILL.md.
Boundary markers (absent): The skill does not employ delimiters or specific instructions to ensure the agent ignores potential commands embedded within the processed novel text.
Capability inventory: The skill provides capabilities for arbitrary file reading via its scripts and performs regex-based text extraction.
Sanitization (absent): The input text is not escaped, filtered, or otherwise sanitized before being included in the output reports and returned to the agent context.

novel-truncator