punk-cover

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content such as articles and topics to generate image prompts, which could potentially contain malicious instructions intended to manipulate the agent's output.
  • Ingestion points: Source material, article text, and user-provided titles are ingested in SKILL.md (Workflow Step 1) to derive visual subjects.
  • Boundary markers: SKILL.md (Workflow Steps 1 and 5) instructs the agent to convert long source material into concise derived fields and context summaries, creating a logical boundary between the raw input and the generated prompt.
  • Capability inventory: The skill uses an image generation tool (image_gen) and has the ability to write prompt files and image artifacts to the punk-assets/ directory as specified in SKILL.md (Steps 6 and 7).
  • Sanitization: The workflow explicitly mandates that the agent must not paste the original article body into the final prompt and instead must use derived fields, providing a layer of protection against direct command injection from the source material.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 01:41 AM
Security Audit — agent-trust-hub — punk-cover