The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the @stripe/link-cli package from the NPM registry. This is an official tool provided by a well-known technology company for managing Link payments.
[COMMAND_EXECUTION]: The skill relies on shell commands to interact with the Stripe Link CLI. This includes managing authentication, retrieving payment methods, and executing payment flows. The instructions use specific flags like --format json to ensure the agent receives structured, predictable output.
[DATA_EXFILTRATION]: The skill manages sensitive payment information (one-time virtual cards). To protect this data, it explicitly instructs the agent to save details to a temporary file (/tmp/link-card.json) with restricted permissions (0600) and forbids the agent from printing or reading the card's PAN into its own context or logs. The instructions include a cleanup step to delete the sensitive file immediately after use.
[INDIRECT_PROMPT_INJECTION]: The skill features a potential attack surface where it ingests WWW-Authenticate headers from external merchant APIs.
Ingestion points: Merchant challenge headers are passed to the link-cli mpp decode command in Section 2 of the procedure.
Boundary markers: The skill does not define specific delimiters for the output of the decode command, but it instructs the agent on how to interpret the results for the next payment step.
Capability inventory: The agent has access to the terminal tool, allowing it to execute CLI commands and manage local files in /tmp.
Sanitization: Validation and decoding of the untrusted merchant data are performed by the underlying @stripe/link-cli tool rather than the LLM, reducing the risk of malicious instructions influencing the agent's logic.

stripe-link-cli