claude-design

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process a wide range of external repository data to inform its design work, creating a vulnerability to indirect prompt injection from malicious files.
  • Ingestion points: The agent is directed to read brand documents, UI kits, theme files, global stylesheets, and component source code from the local filesystem (SKILL.md).
  • Boundary markers: The instructions do not define delimiters or provide guidance to disregard instructions embedded within ingested source files.
  • Capability inventory: The skill utilizes file read capabilities for context gathering and file write capabilities to generate HTML artifacts.
  • Sanitization: Although the skill recommends syntax checks for generated output, it lacks sanitization or validation mechanisms for the input data read from the repository.
  • [METADATA_POISONING]: The frontmatter lists the author as 'BadTechBandit', which contradicts the expected author metadata of 'atlasomnia'. While potentially benign, this represents an inconsistency in skill metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 07:55 AM
Security Audit — agent-trust-hub — claude-design