skill-tester

Warn

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/script_tester.py utility executes Python scripts found in a user-provided directory using subprocess.run. This is the core mechanism for testing script behavior (such as verifying --help output and processing sample data), but it poses a high risk as it runs arbitrary code from the target directory on the host system. The skill documentation explicitly warns that this executes the target's code.
  • [DYNAMIC_EXECUTION]: The skill dynamically discovers and executes Python scripts from the file system at runtime based on a user-provided path. This pattern allows for the execution of content that is not part of the skill's own codebase.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external skill packages (including SKILL.md, README.md, and Python source files) during validation, testing, and scoring. There are no explicit boundary markers or sanitization steps to prevent maliciously crafted content within these files from influencing the AI agent's context or behavior during an audit. The skill's ingestion of untrusted data is paired with high-risk capabilities like command execution and file system access.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 12, 2026, 07:07 PM