The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external GeoJSON data and interpolates properties into HTML popups and tooltips without sufficient sanitization. If an attacker provides a malicious GeoJSON file via a URL or local data, it could lead to Cross-Site Scripting (XSS) when the user opens the generated HTML file.
Ingestion points: add_geojson (accepts arbitrary dictionaries) and add_geojson_url (fetches data from external URLs) in scripts/map_builder.py.
Boundary markers: None identified for external GeoJSON properties.
Capability inventory: The script performs file-system write operations via save_html in scripts/map_builder.py.
Sanitization: While the map title is escaped using _escape_html, feature properties used in bindPopup and bindTooltip are concatenated directly into HTML strings in the generated JavaScript.
[EXTERNAL_DOWNLOADS]: The skill references and fetches resources from several well-known and trusted external services.
Evidence: Fetches historical boundary data from github.com/aourednik/historical-basemaps.
Evidence: Loads marker icons from github.com/pointhi/leaflet-color-markers.
Evidence: Loads the Leaflet.js library and CSS from the unpkg.com CDN.
Evidence: Utilizes map tile services from OpenStreetMap, Stadia Maps, CartoDB, MapWarper, and Academia Sinica (CCTS).
[COMMAND_EXECUTION]: The skill instructions suggest executing shell commands to view the generated output.
Evidence: SKILL.md contains instructions to use the open command to view the generated standalone HTML files in a browser.

historical-map