solidcraft

Warn

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The SKILL.md file contains explicit instructions directing the agent to execute shell commands via osascript to bring the application window to the foreground on macOS.
  • [COMMAND_EXECUTION]: The provided script scripts/sc.py uses subprocess.run() within its wake command to execute osascript commands for system-level window management.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted CAD data and JSON scripts, which serves as a surface for indirect prompt injection attacks.
  • Ingestion points: Untrusted content enters the agent's context through file loading commands such as doc.open, file.insert_step, and solidcraft-cli run.
  • Boundary markers: Absent; the skill instructions do not implement delimiters or ignore-instructions warnings for processed external files.
  • Capability inventory: The skill can perform filesystem writes (file.export), execute CAD engine commands, and invoke system-level shell commands (osascript).
  • Sanitization: Beyond basic JSON schema validation for parameters, the skill does not sanitize the semantic content of processed files.
  • [EXTERNAL_DOWNLOADS]: The skill documentation references and encourages fetching additional modeling resources and functional skills from the tangency-labs GitHub organization.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 17, 2026, 09:01 AM