The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches weather data from the external Open-Meteo API, which represents an ingestion point for untrusted data into the agent's context.
Ingestion points: scripts/fetch_weather.py retrieves real-time weather information from https://api.open-meteo.com/v1/forecast.
Boundary markers: The skill lacks explicit delimiters or instructions to the model to ignore potential malicious content that could be embedded in weather condition strings.
Capability inventory: The skill is restricted to file operations within the .tmp/ directory and network GET requests to a specific weather API. It cannot perform shell execution or external network writes.
Sanitization: The script provides basic sanitization by mapping numeric WMO weather codes to a pre-defined set of internal description strings, reducing the risk of arbitrary string injection from the API.
[METADATA_POISONING]: There is a discrepancy between the data source used in the logic and the attribution provided in the output. While scripts/fetch_weather.py correctly utilizes the Open-Meteo API, the PDF generation script scripts/generate_report_pdf.py hardcodes "OpenWeatherMap" and "openweathermap.org" as the data sources in the report footer and attribution page. This misleading metadata appears to be a developmental oversight rather than a malicious attempt to deceive.

generate-report