The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from Stripe resources, such as customer metadata and product descriptions, which could potentially contain malicious instructions from external sources.
Ingestion points: Untrusted data enters the agent's context via list_customers, search_stripe_resources, and other resource-fetching tools defined in tools.md.
Boundary markers: The skill enforces strict "Hard Rules" in SKILL.md, requiring the agent to display object details, explain consequences, and wait for explicit user confirmation before executing any state-changing or financial operations.
Capability inventory: The skill has access to sensitive billing operations (create_refund, cancel_subscription), as well as Bash and WebFetch tools.
Sanitization: No technical sanitization of metadata strings is described, though the human confirmation gate serves as the primary mitigation.
[EXTERNAL_DOWNLOADS]: The skill references and facilitates connection to official Stripe tools and services.
Connects to the official Stripe MCP server at https://mcp.stripe.com/ for automated operations as described in SETUP.md.
References the official Stripe CLI installation through the Homebrew package manager (stripe/stripe-cli/stripe) and standard Stripe documentation.

stripe