The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents the official installation method via curl -fsSL https://cli.runpod.net | sudo bash. This is a vendor-owned domain and represents the standard, trusted installation path for the runpodctl utility.
[PRIVILEGE_ESCALATION]: The use of sudo is limited to the installation documentation for global binary placement, which is typical for CLI tools and does not indicate an attempt to exploit the system.
[INDIRECT_PROMPT_INJECTION]: The skill manages untrusted data ingestion from the Runpod Hub and Hugging Face (SKILL.md). It provides clear boundary markers by specifying that output is JSON-formatted and should be parsed accordingly (reference/output-and-errors.md). The capability inventory includes pod and serverless resource management via Bash(runpodctl:*). Sanitization is addressed by instructing the agent to use structured parsing tools like jq and to handle coded error responses rather than free-text messages.
[CREDENTIALS_UNSAFE]: The skill actively discourages unsafe credential handling. It provides instructions for piping registry tokens via stdin to prevent exposure in process tables and warns against sharing worker logs that might contain internal API keys.

runpodctl