The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external book files, which may contain adversarial instructions designed to influence the agent's summary or distillation logic.
Ingestion points: SKILL.md identifies EPUB, PDF, text, Markdown, HTML, and DOCX files as supported inputs for distillation.
Boundary markers: The workflow in SKILL.md defines specific labeling rules for "Author's claim", "Direct quote", and "Analysis" to help distinguish source content from AI reasoning.
Capability inventory: The agent can execute local Python scripts (scripts/extract_epub.py and scripts/build_html_report.py) to process files and write output.
Sanitization: The extract_epub.py script uses html.parser and regex to clean extracted text. The build_html_report.py script applies HTML escaping to content before rendering the final interactive reader, though it retains a potential surface for link-based XSS if the agent is induced to generate malicious URLs in the Markdown report.
[COMMAND_EXECUTION]: The skill uses shell commands to invoke local Python scripts for document parsing and report generation.
Evidence: SKILL.md workflow steps 2 and 6 execute python3 scripts/extract_epub.py and python3 scripts/build_html_report.py. These scripts use only standard library modules and do not perform network operations or download remote code.

distill-book