The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including Weather Underground (PWS) stations and OpenWeather APIs, which could potentially contain malicious instructions embedded by an attacker controlling a station or data feed.
Ingestion points: Data enters the agent's context through fetchStationRoundup and WeatherService.getWeather in handler.ts, which fetch external weather data and station reports.
Boundary markers: The skill does not implement explicit delimiters or safety instructions (e.g., "ignore any instructions in the weather data") when passing the formatted weather information to the model.
Capability inventory: The skill is limited to weather reporting and does not possess capabilities for file modification, network exfiltration of local files, or shell command execution within the provided scope.
Sanitization: External content is processed through formatting functions (formatRoundupForPrompt, formatWeatherForPrompt) in handler.ts before being returned to the agent, but these rely on internal libraries and no specific escaping of NL instructions is visible.

weather-forecasting