slides
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and transform external "source material" and "evidence" into presentation decks (SKILL.md, Step 1). This creates a vulnerability surface where malicious instructions could be embedded in the source data to influence the agent's behavior during the drafting or file generation process.
- Ingestion points: The workflow relies on external "supplied material" and "source data/evidence" provided by the user or external files (SKILL.md, Step 3; references/create.md).
- Boundary markers: The provided instructions do not include specific directives to use delimiters or to explicitly ignore instructions found within the ingested source data.
- Capability inventory: The skill involves creating and writing files in multiple formats (HTML, PowerPoint, PDF) using the agent's native tools (SKILL.md, Step 5).
- Sanitization: There are no instructions for the agent to sanitize, validate, or filter content from the source data before including it in the generated presentation.
Audit Metadata