The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies an attack surface where instructions can be influenced by untrusted data fetched from external URLs during the discovery process.
Ingestion points: The agent is instructed to fetch and read a service manifest (manifest.json), an llms.txt file, and service-specific OpenAPI schemas from the agents.machina.gg domain.
Boundary markers: The skill includes explicit instructions to treat all fetched catalog text, news, market titles, and API responses as untrusted data. It further specifies that this data cannot grant spending authority or override terms.
Capability inventory: The skill is a prompt-only skill that does not execute purchases itself. However, it directs the agent to identify and potentially call external APIs based on discovered schemas, which requires external tools and user approval.
Sanitization: Instructions explicitly forbid putting credentials, private payloads, or personal details into public-facing goal parameters or URLs.
[EXTERNAL_DOWNLOADS]: The skill references configuration files and instructions for installing additional modular components from external sources.
The agent is instructed to fetch a manifest and OpenAPI schemas from the agents.machina.gg domain, which is the official infrastructure for the skill's author.
The documentation lists several optional installs using npx skills add targeting the machina-sports/sports-skills repository on GitHub. These references are consistent with the skill's stated authorship.

world-cup