The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a predefined set of tools (e.g., list-pods, list-billing) for data retrieval. These tools are restricted to information gathering and do not allow for arbitrary command execution.
[DATA_EXFILTRATION]: While the skill accesses billing information, this data is retrieved using vendor-provided tools for the purpose of presenting it directly to the user. No unauthorized network operations or exfiltration patterns were detected.
[PROMPT_INJECTION]: The instructions include strong directives for the agent to remain in a read-only state and to defer any resource modifications to other specialized skills, preventing unauthorized actions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from MCP tool outputs. However, because the skill has no destructive capabilities or write access, the risk of an indirect injection attack being used to compromise the system or data is negligible.

cost-audit