epub
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted third-party EPUB files, presenting an untrusted data surface that enters the agent's context.
- Ingestion points: Document ingestion happens in
scripts/epub-reader/src/index.tsvialoadEpub()andgetChapterContent(), where chapter content is read and returned as plain text. - Boundary markers: No explicit XML/markdown boundaries or instructions to ignore instructions are wrapped around the extracted text output forwarded to the agent.
- Capability inventory: The tool's capabilities are strictly limited to local file extraction, metadata reading, table-of-contents formatting, and substring searching. No shell execution, writing of files, or network operations are exposed.
- Sanitization: HTML text content is converted into Markdown via
turndown, but no content filtering is performed to look for malicious prompt injection instructions embedded inside book contents.
Audit Metadata