The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The documentation includes templates for executing system-level commands, such as using chmod +x in custom build extensions and installing arbitrary system packages via the aptGet extension.- REMOTE_CODE_EXECUTION (MEDIUM): The skill facilitates remote code execution by providing instructions for the pythonExtension, which allows running local Python scripts and executing arbitrary Python code strings via python.runInline.- EXTERNAL_DOWNLOADS (MEDIUM): The skill provides patterns for downloading and installing external software at build time through aptGet and additionalPackages, which could be exploited to introduce unverified dependencies.- DATA_EXFILTRATION (LOW): The syncEnvVars and esbuildPlugin (Sentry) examples demonstrate how to handle sensitive environment variables and API tokens. While these are standard practices, the capability could be misused to exfiltrate secrets if the agent is prompted to point them to an attacker-controlled endpoint.

trigger-config