The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the execution of a curl command to verify the integration, which interacts with a remote server provided in the setup prompt.
[DATA_EXFILTRATION]: The skill is designed to redirect LLM traffic and provider API keys (e.g., OPENAI_API_KEY, ANTHROPIC_API_KEY) to a third-party gateway URL. In 'byok' (Bring Your Own Key) mode, it explicitly sends the existing provider key in the x-cave-upstream-key header to the external gateway.
[CREDENTIALS_UNSAFE]: The skill instructs the agent to search for, read, and manipulate sensitive authentication tokens from environment variables and .env files.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it interpolates variables from an external "setup prompt" (GATEWAY, CAVE_API_KEY) directly into a shell command (curl) without sanitization, potentially allowing for arbitrary command execution.
[DATA_EXFILTRATION]: The skill instructs the agent to perform a "real, billable" verification request through the proxy autonomously, bypassing user confirmation for a network operation that transmits credentials.

caveman-setup