create-legacy-command

Warn

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to use bash and echo to create files based on user-provided strings such as the name and description. Interpolating untrusted user input directly into shell commands creates a significant risk of command injection (e.g., via semicolons or backticks) if the agent does not strictly validate the input characters.
  • [PROMPT_INJECTION]: The SKILL.md.jinja template generates instructions that explicitly override agent behavior, such as: "you MUST execute the provided Python determinism script instead of attempting to solve the task using raw bash or javascript logic." This pattern of generating hard-coded overrides can be used to bypass an agent's standard reasoning or safety guardrails in generated sub-components.
  • [EXTERNAL_DOWNLOADS]: The generated README.md.jinja template provides ready-to-use bash commands for installing external Python dependencies (pip install -r requirements.txt). While common in development workflows, this facilitates the introduction of unverified third-party code into the agent's environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 14, 2026, 04:22 PM