The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates as a content transformation layer, ingesting untrusted data (URLs, files, raw text) via the input parameter. This presents a vulnerability surface for indirect prompt injection.\n
Ingestion points: The input parameter defined in SKILL.md accepts external URLs and file data URI payloads.\n
Boundary markers: The skill explicitly instructs agents to treat converted content as untrusted data and ignore any embedded directives or instructions within the output.\n
Capability inventory: The skill utilizes network tools to interact with the mdapi.io API and includes logic for handling and signing autonomous payments.\n
Sanitization: The instructions emphasize that transformed output should be treated strictly as data and never as a source of instructions for subsequent agent actions such as payments or data exfiltration.\n- [EXTERNAL_DOWNLOADS]: The skill contains instructions to fetch its own updated definitions and various discovery manifests from the vendor's official domain.\n
Evidence: The Skill freshness check section directs agents to GET https://mdapi.io/.well-known/skill.md to retrieve the latest reference. Additionally, manifests like ai-discovery.json, agent.json, and acp.json are referenced for protocol discovery. These downloads are targeted at the author's verified domain and are used for legitimate configuration and discovery purposes.

mdapi-conversion