The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents an integration feature, askAssistant, which enables Hosty-managed applications to programmatically send text to the AI agent's context.
Ingestion points: Data enters the agent's context through the askAssistant function provided by the @hosty-sdk/app library, as described in references/app-auth-and-users.md.
Boundary markers: The protocol is limited to plain-text input and requires the host shell to handle the handoff (e.g., preparing a draft) rather than executing instructions immediately.
Capability inventory: The documentation explicitly states that authentication and tool permissions remain active and governed by the host, preventing the injected text from bypassing existing security controls.
Sanitization: The host platform enforces a 4000-character limit and performs mandatory sender origin verification to ensure the message comes from a valid application frame.
[COMMAND_EXECUTION]: The application manifest contract allows for defining lifecycle commands used to prepare and run services within the Hosty environment.
Evidence: The setup field in the manifest (documented in references/app-manifest.md and references/app-dev-mode.md) allows for standard initialization commands such as npm install, dotnet restore, or pip install to prepare source checkouts.
[EXTERNAL_DOWNLOADS]: The platform's application discovery and installation mechanism facilitates fetching manifest files and update feeds from remote sources.
Evidence: references/app-feeds.md describes the feeds.json contract, which utilizes manifestRef URLs to retrieve application metadata from external hosting services like GitHub.

hosty-app-skill