The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted data from web origins and JavaScript bridges, which is a known attack surface for indirect prompt injection.
Ingestion points: Web content loaded into webview_flutter or similar components, specifically through JavaScript channels and navigation requests as described in SKILL.md and references/javascript-bridges-and-permissions.md.
Boundary markers: The skill explicitly instructs the agent to use strict boundary markers, such as parsing every navigation as a URI instead of a string prefix and validating every JavaScript payload against a typed Dart contract.
Capability inventory: The skill provides guidance on minimizing capabilities, explicitly warning against exposing the filesystem, platform channels, or identity mutation to the WebView runtime.
Sanitization: The instructions mandate parsing bounded JSON, rejecting unknown fields, and validating the origin of every message in Dart.
[EXTERNAL_DOWNLOADS]: The skill references and directs the agent to utilize official resources from trusted organizations and well-known services.
Evidence: References to official documentation and packages are provided for pub.dev (Google), developer.android.com (Google), developer.apple.com (Apple), and rfc-editor.org.
[COMMAND_EXECUTION]: The skill contains no executable code or shell command patterns. It serves as a set of instructional constraints for secure development.

flutter-webview