The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources such as account audit logs, organization member lists, and resource metadata (names and descriptions) which could be controlled by other users or processes. This creates a surface for indirect prompt injection where malicious instructions embedded in these logs or names could influence the agent during resource management operations.
Ingestion points: Data retrieved from list-templates, list-audit-logs, and list-organization-users as described in SKILL.md.
Boundary markers: None identified in the prompt templates or instructions.
Capability inventory: The skill has extensive mutation capabilities including update-ssh-keys, create-cluster, create-secret, and delete-network-volume across several tool-binding sections in SKILL.md.
Sanitization: No explicit sanitization or filtering of input from read tools before using it in mutation tool arguments is specified.
[COMMAND_EXECUTION]: The skill facilitates complex command execution through a large suite of Runpod MCP tools. It handles sensitive operations such as managing SSH public keys and creating billable GPU clusters. While it includes strong instructional guardrails—such as requiring cost confirmation for billable resources and prohibiting the echoing of secret values—the inherent power of the available tools presents a high-capability surface area.

runpod-lifecycle-crud