character-casting-studio
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user-supplied character briefs—which may include descriptions of physical appearance, clothing, and background—and transform them into structured prompts for an image generation tool. This interaction model presents a surface for indirect prompt injection where a user might attempt to bypass the skill's constraints (e.g., requesting prohibited celebrity likenesses).
- Ingestion points: User input through natural language character descriptions in
SKILL.mdandoriginal-spec.md. - Boundary markers: The skill structures ingestion through defined fields (identity, face, distinctive details, hair/makeup, wardrobe) which acts as a logical barrier, though it does not use explicit character-based delimiters like XML tags.
- Capability inventory: The skill is intended to interact with an image generation tool (e.g., DALL-E) to produce character casting sheets.
- Sanitization: The instructions provide a robust set of negative constraints (e.g., 'no celebrity likeness', 'no recognizable existing character') and mandatory quality gates that the agent must verify before outputting results, effectively acting as a prompt-level filter for user-provided data.
Audit Metadata