The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for rendering external GeoJSON data directly into the map's UI via popups and tooltips. If the GeoJSON source is attacker-controlled, malicious scripts or instructions could be injected into the agent's context or the user's browser.
Ingestion points: Untrusted data enters the context through L.geoJSON(geojsonData, ...) and incremental data loading via geoLayer.addData() as seen in SKILL.md and examples/core.md.
Boundary markers: The examples do not include boundary markers or warnings to ignore embedded instructions within the ingested GeoJSON properties.
Capability inventory: The skill uses bindPopup(), bindTooltip(), and L.divIcon(), which accept and render HTML strings into the DOM.
Sanitization: The provided code snippets (e.g., layer.bindPopup(feature.properties?.name)) interpolate data properties directly into HTML-rendering methods without escaping or sanitization.
[EXTERNAL_DOWNLOADS]: The skill references several external tile providers for map rendering.
Evidence: Fetches map tiles from tile.openstreetmap.org (OpenStreetMap) and basemaps.cartocdn.com (CARTO).
Context: These are well-known, industry-standard services for web mapping and are used here for their intended purpose. The references do not include sensitive data exfiltration.

web-maps-leaflet