The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes XHTML content from external EPUB files and passes it to LLMs for knowledge extraction in scripts/epub-extract-knowledge. The extraction logic lacks robust delimiters or explicit instructions for the LLM to ignore potentially malicious instructions embedded within the ebook text, creating a surface for indirect prompt injection attacks.
[COMMAND_EXECUTION]: Several scripts, including scripts/epub-batch and scripts/epub-validate, utilize subprocess.run to execute sibling scripts or external system tools like Java and EPUBCheck. While these operations are constrained to specific tasks and internal tools, they represent a capability for system command execution that could be targeted if the agent's context is compromised.
[DATA_EXFILTRATION]: The scripts/epub-extract-knowledge script facilitates network communication by sending extracted ebook text to a user-provided LLM endpoint via HTTP POST requests. Although this is a core feature for analysis, it enables data transmission to external, non-whitelisted domains defined via environment variables.

epub