tldr-prompt
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from external URLs and local workspace files to create summaries, presenting an indirect prompt injection surface.
- Ingestion points: Content is retrieved from user-specified or automatically resolved URLs via
#tool:fetch, as well as from local workspace configuration files (.prompt.md,.agent.md, etc.). - Boundary markers: Absent. The instructions lack explicit constraints or delimiters instructing the agent to ignore any natural language instructions embedded within the fetched documents.
- Capability inventory: The skill is restricted to reading operations and chat output generation. It does not possess file-writing, system command execution, or elevated privileges.
- Sanitization: Absent. Retextualization and template-fitting are performed directly on raw input text without escaping or content sanitization mechanisms.
Audit Metadata