editorial-portrait-composition

Pass

Audited by Gen Agent Trust Hub on Aug 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it requires reading and processing external data from references/cases.md (Category 8).
  • Ingestion points: The agent is explicitly instructed to read references/cases.md on every invocation to select an 'anchor Case'.
  • Boundary markers: The instructions include a 'Required evidence gate' and safety sections to prevent 'superficial copying' and 'impersonation', acting as soft boundaries.
  • Capability inventory: The skill uses image/browser tools to inspect media and generates text-based image prompts; it does not have direct file-write or system-level execution capabilities.
  • Sanitization: There is no explicit technical sanitization (like JSON schema validation) of the external case data beyond the LLM's internal filters and the 'Preserve / Replace / Avoid' logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 23, 2026, 06:36 AM
Security Audit — agent-trust-hub — editorial-portrait-composition