superdesign

Warn

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches its core initialization instructions (INIT.md) and operational guidelines (SUPERDESIGN.md) from a remote GitHub repository at runtime.\n- [COMMAND_EXECUTION]: The skill mandates the execution of shell commands, including npm install -g @superdesign/cli@latest and superdesign login, and instructs the agent to perform these actions automatically without manual user intervention.\n- [REMOTE_CODE_EXECUTION]: By requiring the installation of a global NPM package from an external source and subsequently running it, the skill introduces a mechanism for code execution on the host system.\n- [DATA_EXFILTRATION]: The skill performs an automated scan of the local repository, extracting "FULL source code" for components, layouts, and theme configurations into a local directory (.superdesign/init/), which is then processed by the external CLI.\n- [CREDENTIALS_UNSAFE]: The skill instructs the agent to run a login command using a third-party CLI tool, which introduces potential risks related to the handling and exposure of user credentials.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes project source code and configuration files as untrusted context for design tasks, creating a surface for indirect prompt injection.\n
  • Ingestion points: Reads package.json, source files (.tsx, .ts, .css), and framework configuration files (e.g., next.config.js) to build context.\n
  • Boundary markers: None identified; raw code implementation is directly included in context files.\n
  • Capability inventory: Executes shell commands (npm install, superdesign), performs file writes to the .superdesign/ directory, and utilizes the CLI which likely performs network operations.\n
  • Sanitization: The skill does not perform validation, filtering, or escaping of the code snippets ingested from the repository.\n- [METADATA_POISONING]: There is a discrepancy between the author declared in the skill's metadata (superdesign) and the author identity recognized by the platform (yasikvlad), which may mislead users about the skill's provenance.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 15, 2026, 07:14 PM