creative-research-first
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core functionality relies on ingesting and processing untrusted external data from web research, creating a surface for indirect prompt injection attacks.
- Ingestion points: The workflow involves conducting web searches based on research questions and extracting content from museum, academic, and community sources (SKILL.md, references/research-playbook.md).
- Boundary markers: The skill uses an "Extraction Card" and "Research Lanes" to structure data (SKILL.md), but it does not provide explicit technical delimiters (e.g., XML tags or specific block quotes) or instructions for the agent to ignore potentially malicious embedded commands within retrieved search snippets.
- Capability inventory: The synthesized output is a "complete prompt" (SKILL.md) intended for further creative generation, which could influence downstream agent actions or tool executions depending on the broader environment.
- Sanitization: The instructions include behavioral guardrails such as "do not copy... verbatim" and "translate findings into original design controls" (SKILL.md), which act as a manual sanitization layer during the creative synthesis process.
Audit Metadata