The Agent Skills Directory

Interaction with Untrusted Merchant Content: The skill processes content from merchant websites, such as DOM elements and HTTP headers, to facilitate checkout. While this represents a potential surface for indirect prompt injection, the skill includes explicit instructions to treat all merchant-controlled content as untrusted data and to avoid following directives embedded within it.
Handling of Sensitive Payment Information: The skill manages high-value data including virtual card numbers, tokens, and personally identifiable information (PII). It incorporates safety practices like recommending the use of specific file permissions (0600) for credential storage and advising the agent to mask sensitive details when displaying them to the user.
Command Execution and Argument Safety: The skill uses the link-cli tool for transaction workflows. It provides guidance on using structured command arguments (_next.pay_argv) to avoid potential shell injection issues when processing data returned from merchant APIs.
Ingestion of External Data: The agent is instructed to browse merchant pages and interpret challenges (e.g., HTTP 402 errors). The skill documentation emphasizes that these values must be treated as data even when passed through CLI continuations, providing a layer of defense against malicious merchant responses.

create-payment-credential