resourceful-problem-solving
Warn
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses 'always: true' and explicit instructions to 'Never refuse a task' and 'Attempt a code-based solution first,' which are designed to override the agent's default safety guardrails and refusal protocols.
- [EXTERNAL_DOWNLOADS]: The instructions direct the agent to install arbitrary, unversioned software packages and CLI tools from public registries like PyPI (pip), npm, and Homebrew (brew) based on user requests.
- [COMMAND_EXECUTION]: The skill encourages the agent to execute various shell commands for searching and installing software (e.g., 'brew search', 'pip install', 'npm install') and running generated scripts.
- [REMOTE_CODE_EXECUTION]: By instructing the agent to dynamically write and execute scripts that download and run third-party dependencies from the internet, the skill establishes a vector for remote code execution via untrusted packages.
Audit Metadata