compose-preview-design-board

Fail

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill recommends a bootstrap installation method that pipes a remote script from the author's repository directly into the shell.
  • Evidence: curl -fsSL https://raw.githubusercontent.com/yschimke/compose-ai-tools/main/scripts/install.sh | bash in SKILL.md.
  • [EXTERNAL_DOWNLOADS]: The skill retrieves an installer from an external GitHub repository controlled by the author.
  • [COMMAND_EXECUTION]: The instructions involve executing local commands and a Python utility to manage files and generate output.
  • [PROMPT_INJECTION]: The scripts/build-design-board.py script presents an attack surface for indirect prompt injection via its processing of untrusted JSON specifications.
  • Ingestion points: The script parses JSON input from a file or standard input in scripts/build-design-board.py.
  • Boundary markers: No boundary markers are used to isolate user-controlled data within the JSON spec.
  • Capability inventory: The script can read arbitrary files from the filesystem and include their contents as base64 in the generated HTML output.
  • Sanitization: While HTML special characters are escaped in the output, the script does not validate that the requested file paths are restricted to the intended project directories.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/yschimke/compose-ai-tools/main/scripts/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 24, 2026, 01:27 PM
Security Audit — agent-trust-hub — compose-preview-design-board