cover-image

Warn

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses a shell command to execute a local Python script: uv run ~/.claude/skills/_generate_image.py "<approved prompt>". This pattern is vulnerable to command injection if the variable, which is derived from user input, contains shell metacharacters like backticks or semicolons.\n- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes untrusted user articles into command arguments.\n
  • Ingestion points: Articles provided by the user in the first step of the workflow.\n
  • Boundary markers: None present to delimit untrusted article content from the prompt generation logic.\n
  • Capability inventory: Uses the Bash tool to execute subprocesses via uv.\n
  • Sanitization: No automated sanitization of the generated prompt is described; it relies on manual user approval which may not catch technical injection attacks or obfuscated shell commands.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 4, 2026, 04:13 PM