cadcraft-layer-cleanup

Warn

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches a portable command-line tool from an external GitHub release.
  • Evidence: https://github.com/storytold/cadcraft/releases/download/v0.7.0/cadcraft-0.7.0-windows-x64-portable.zip contains the cadcraft-cli.exe binary.
  • The skill also references external CI run logs at https://github.com/partme-ai/full-aigc-plugins/actions/runs/38921194883.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes entity data from CAD drawings, which serves as an ingestion point for untrusted content.
  • Ingestion points: Reading entity metadata and geometry via drawing.inspect in SKILL.md and references/craft-methods.md.
  • Boundary markers: None identified to distinguish between drawing data and agent instructions.
  • Capability inventory: The skill has the ability to write to the file system (save), execute CAD-specific commands (execute), and install additional software packages (npx skills add).
  • Sanitization: The instructions recommend using argv for CLI operations to mitigate shell-related quoting issues, but no sanitization is mentioned for the drawing data itself.
  • [COMMAND_EXECUTION]: The skill provides instructions for executing shell commands and binary tools.
  • Evidence: Instructions for npx skills add full-aigc-skills/cadcraft-skills and running the cadcraft-cli.exe tool with arbitrary command arguments (--cmd).
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 4, 2026, 06:48 PM