designly-director
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing untrusted user input (e.g., '{{BRIEF}}', '{{REFERENCES}}') to orchestrate complex design workflows. This creates a potential surface where malicious instructions embedded in a design brief could attempt to influence the agent's orchestration logic or the specialized downstream skills it coordinates.\n
- Ingestion points: Multiple placeholders in
references/prompt-playground.md(Workflows WF-01 through WF-16) and intent interpretation logic inSKILL.md.\n - Boundary markers: The skill uses structural headers like 'INPUTS' and 'OPERATING INSTRUCTIONS' to separate data from logic, but lacks explicit guardrails or sanitization instructions to ignore adversarial commands within the ingested text.\n
- Capability inventory: The orchestrator controls a suite of 21 specialized skills for image generation, editing, and creative strategy, providing a significant capability surface.\n
- Sanitization: While 'edit-sanitizer' is mentioned for image modifications, no specific natural language sanitization is defined for the initial brief interpretation phase.
Audit Metadata