The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data, such as match news and full article bodies, which could contain hidden instructions aimed at influencing the agent's behavior.
Ingestion points: Data retrieved from api.crawlora.net via scripts/crawlora.sh, specifically through news-related endpoints described in reference/endpoints.md (e.g., /fotmob/news-article, /fotmob/latest-news, /fotmob/news).
Boundary markers: The instructions in SKILL.md lack specific guidance for the agent to use delimiters or "ignore" instructions embedded within the fetched football articles.
Capability inventory: The skill provides access to a bash script (scripts/crawlora.sh) that can perform GET requests to a wide range of football data endpoints.
Sanitization: While the script correctly encodes query parameters to prevent command injection, the resulting news text is processed by the agent without specific sanitization for prompt injection patterns.
[SAFE]: The script scripts/crawlora.sh implements secure handling of the CRAWLORA_API_KEY by using umask 077, creating a restricted temporary file for the curl configuration, and verifying the key's character set to prevent injection into the configuration file.
[SAFE]: The skill restricts network communication to the hardcoded vendor domain api.crawlora.net and enforces a strict whitelist of specific API routes in scripts/crawlora.sh, preventing unauthorized access to other API surfaces.

fotmob-research