The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/curl.sh acts as a wrapper for the curl command. It takes user-provided URLs and arguments, modifies them by replacing placeholders (_DOMAIN_ and _TIME_), and executes them. The domain name weather.com.cn is obfuscated using Base64 encoding (d2VhdGhlci5jb20uY24=) within the script, which is a technique used to hide the destination of network requests.
[DATA_EXFILTRATION]: The skill executes network operations to subdomains of a hidden domain. While the target domain weather.com.cn is a well-known meteorological service, the script's mechanism of hiding the domain and dynamically constructing command arguments allows for potential communication with external hosts.
[SAFE]: The identified network targets are official subdomains of the weather service for China, which aligns with the skill's stated purpose of providing weather forecasts.
[PROMPT_INJECTION]:
Ingestion points: External weather data is fetched from subdomains of weather.com.cn via scripts/curl.sh and processed by the agent to display forecasts.
Boundary markers: None identified. There are no instructions to the agent to treat the fetched data as untrusted or to isolate it from the instruction context.
Capability inventory: The skill possesses the capability to execute shell commands and perform network requests through the scripts/curl.sh wrapper script.
Sanitization: There is no evidence of validation, escaping, or sanitization of the data returned from the weather APIs before it enters the agent's context.

weather