The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external and untrusted content from URLs, documents, and images.
Ingestion points: The input parameter in SKILL.md accepts external URLs and data URIs for transformation.
Boundary markers: The instructions explicitly warn the agent to treat all converted content as untrusted data and to ignore any instructions, phishing prompts, or payment scams found within the output.
Capability inventory: The skill performs network operations to the vendor's API and handles payment-related headers.
Sanitization: The skill relies on specific prompt-based instructions to ensure the agent does not execute content found in the transformed output.
[EXTERNAL_DOWNLOADS]: The skill references and fetches resources from the vendor's domain.
Reference fetching: It includes instructions to fetch updated skill definitions from https://mdapi.io/.well-known/skill.md.
QR Code retrieval: The manual payment workflow fetches SVG images from https://mdapi.io/qr.
Context: All external requests target the vendor's primary domain (mdapi.io), which is consistent with the skill's stated purpose.
[COMMAND_EXECUTION]: The autonomous payment flow requires the agent to interact with wallets to sign payment payloads.
Evidence: The skill describes a process where the agent must prepare a payment signature based on a challenge and submit it via the PAYMENT-SIGNATURE header during a request retry.

mdapi-conversion