storytelling
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input such as project context and artifacts, creating an attack surface for indirect prompt injection.
- Ingestion points: The Standalone workflow in SKILL.md specifies that the agent reads user-provided project context and existing design artifacts.
- Boundary markers: SKILL.md contains a Stance and a set of Five Refusals that act as behavioral boundaries to prevent the agent from following instructions that would lead to data smoothing or manufactured tension.
- Capability inventory: All logic within SKILL.md is focused on text generation and narrative structuring; there are no subprocess calls, network requests, or file-writing capabilities present.
- Sanitization: The skill employs behavioral sanitization by directing the agent to explicitly identify missing evidence and surface gaps rather than fabricating information.
Audit Metadata