The Agent Skills Directory

[DATA_EXFILTRATION]: Network communication is performed using curl to a non-whitelisted external domain (https://ru7m5svay1.execute-api.eu-central-1.amazonaws.com). While the script only transmits league and team filters rather than sensitive system files or credentials, communicating with unverified external endpoints poses a minor data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection and schema confusion due to the unvalidated handling of external inputs and remote API responses.
Ingestion points: User-supplied filters (league, home_team, away_team) and remote data returned from the AWS API Gateway endpoint enter the agent's context within scripts/footballbin.sh.
Boundary markers: Absent. The script processes and prints the API response directly to stdout using jq without any encapsulating delimiters or instructions to prevent the agent from executing instructions embedded in the payload.
Capability inventory: The script relies on curl for network requests and jq for parsing data, but it does not execute downstream commands, perform file system writes, or invoke dynamic execution utilities on the data received.
Sanitization: Input variables are interpolated directly into a raw JSON string template ("home_team":"$home_team") without escaping special characters. If a parameter contains malicious double quotes or backslashes, it can break the JSON syntax or alter the payload structure.

footballbin-predictions