The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents several APIs that ingest untrusted data from external sources, creating an attack surface for indirect prompt injection. Ingestion points include references/networking/http-client.md (HTTP content), references/networking/syndication.md (RSS/Atom feeds), references/windows-integration/clipboard.md (system clipboard content), and references/windows-integration/protocol-activation.md (URI parameters). While boundary markers are generally absent from code examples, the protocol-activation.md file contains explicit warnings to treat URI parameters as untrusted input and to validate/sanitize them before use. The documented capabilities include high-privilege access such as internetClient, privateNetworkClientServer, and runFullTrust.
[COMMAND_EXECUTION]: The documentation includes guidance on executing system commands and launching external content via the Launcher and StartupTask APIs. references/windows-integration/launcher.md describes launching files and URIs using default system handlers, while references/windows-integration/startup-task.md provides instructions for registering applications to run automatically at system startup, which is a standard method for establishing persistence on Windows systems.

windows-platform-integration