The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like npm install, npx skills, and the agent-cards CLI to install dependencies and manage the service environment.
[EXTERNAL_DOWNLOADS]: It downloads software from the npm registry (agent-cards) and connects to a remote third-party MCP server (mcp.agentcard.sh) to facilitate card management and payments.
[CREDENTIALS_UNSAFE]: The setup instructions guide the agent to read a sensitive authentication token (JWT) from a local configuration file located at ~/.agent-cards/config.json.
[DATA_EXFILTRATION]: The skill provides tools to retrieve and display highly sensitive financial information, including full card numbers (PAN), CVV codes, expiry dates, and transaction histories, which are transmitted to a remote service endpoint.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Ingestion points: Untrusted data enters the agent context via list_transactions (merchant names) and read_support_chat (message history). Boundary markers: Absent; there are no specific delimiters or warnings to ignore instructions within these data fields. Capability inventory: The agent has high-risk capabilities including create_card, close_card, get_card_details, and pay_checkout. Sanitization: Absent; the skill does not specify any validation or filtering of content retrieved from the external MCP tools.

agent-card