doc-coauthoring
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external sources, creating an attack surface where embedded instructions could override intended behavior.
- Ingestion points: Stage 1 of the workflow in SKILL.md describes fetching data from user 'info dumps', shared document links, and integrated messaging and storage apps like Slack, Teams, Google Drive, and SharePoint.
- Boundary markers: The instructions lack specific delimiters or system-level guidance to ignore embedded prompts within the external text or files being analyzed.
- Capability inventory: The skill utilizes tools for file creation and modification (create_file, str_replace) and describes a 'Reader Testing' phase that involves invoking sub-agents to process the document content.
- Sanitization: No validation or sanitization routines are implemented to check external data for prompt injection patterns before the content is passed to the LLM or sub-agents.
Audit Metadata