The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for a fallback installation method that involves downloading a shell script from the vendor's GitHub repository and executing it using bash after an optional inspection step.
Evidence: curl -fsSL https://raw.githubusercontent.com/machina-sports/machina-cli/main/install.sh -o /tmp/machina-install.sh followed by bash /tmp/machina-install.sh in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent or user to install the machina-cli tool using package managers or by downloading a script from a remote GitHub URL.
Evidence: pipx install machina-cli and curl commands targeting raw.githubusercontent.com/machina-sports/machina-cli/ in SKILL.md.
[COMMAND_EXECUTION]: The skill relies on the machina-cli binary to perform various operations including login, project selection, and authentication checks via the shell.
Evidence: Commands such as machina login, machina project use <id>, and machina auth whoami are documented for use in SKILL.md.
[PROMPT_INJECTION]: The skill retrieves and processes data from external, untrusted sources including prediction markets and social media sentiment, presenting a risk of indirect prompt injection.
Ingestion points: External data is ingested through tools such as worldcup-get-market-state (market data) and worldcup-fan-sentiment-context (social media pulse) as described in SKILL.md.
Boundary markers: The instructions do not implement specific delimiters or warnings to the agent to disregard instructions embedded in the external content.
Capability inventory: The skill environment includes the ability to execute shell commands via the machina-cli tool (documented in SKILL.md).
Sanitization: There is no mention of sanitization, filtering, or validation processes for the data retrieved from external venues.

world-cup