canvas-design

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [PROMPT_INJECTION]: The instructions utilize simulated user feedback (e.g., 'The user ALREADY said "It isn't perfect enough. It must be pristine..."') to pre-emptively command the agent into a specific, high-performance artistic mode. While intended for quality control, this mimics a common injection technique used to override or constrain agent behavior.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied creative concepts to generate a design philosophy, which creates an attack surface for indirect instructions.\n
  • Ingestion points: User-provided creative direction and subtle inputs (described in SKILL.md).\n
  • Boundary markers: Absent; there are no delimiters or specific instructions to disregard embedded commands in user-provided concepts.\n
  • Capability inventory: The agent is authorized to generate Markdown files and binary visual files (.pdf, .png), and is instructed to search the local './canvas-fonts' directory.\n
  • Sanitization: Absent; concepts are directly interpreted and expressed visually or through descriptive text without validation or filtering.\n- [METADATA_POISONING]: A metadata discrepancy exists between the YAML frontmatter, which specifies an 'MIT License', and the 'LICENSE.txt' file, which contains the Apache License 2.0 text. This is a best practice violation regarding metadata accuracy.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 12:41 PM
Security Audit — agent-trust-hub — canvas-design