export declare const AN_SKILL_MD = "---\nname: an\ndescription: >-\n  Open and operate Agent-Native workspace apps through Dispatch MCP, with\n  inline app surfaces, browser-session authentication, and current screen state.\nmetadata:\n  visibility: exported\n---\n\n# Agent-Native Apps\n\nUse `/an <app>` to open an Agent-Native app beside the conversation and keep\nthe app's live UI and the agent's actions in the same workflow. The first\nargument is the Dispatch app id. For example, `/an slides` opens the Slides\neditor. After it opens, the user can sign in in that browser surface and ask\nfor new decks or focused edits in ordinary language.\n\n## Open the app\n\nFor a known app, call Dispatch's `open_app` with an inline surface:\n\n```json\n{\n  \"app\": \"slides\",\n  \"path\": \"/\",\n  \"embed\": true,\n  \"chrome\": \"full\"\n}\n```\n\nThe MCP App response is the preferred handoff for Claude, Claude Desktop,\nClaude Cowork, ChatGPT, ChatGPT desktop, Codex, and other compatible hosts. Keep\nthe returned open link as the fallback for a host that cannot render MCP Apps\ninline. Do not claim that a browser opened unless the host rendered the inline\nsurface or its browser tool confirmed the returned URL.\n\nIf the app name is missing or uncertain, call `list_apps` first and use the\nexact granted app id. If the requested app is not listed, explain that it must\nbe granted in Dispatch's Agents access settings. Do not invent a URL, use a\nlocalhost URL, or route around the Dispatch grant.\n\n## Work in the opened app\n\nPrefer direct app tools when the host exposes them. After opening an app, call\n`list-host-webmcp-tools` when available, then use `run-host-webmcp-tool`\nwith the exact listed name and origin. Page-local tools reflect current app\nstate and selection, so use `view-screen` before selection-dependent edits\nand read back writes.\n\nYou are the model doing the work: when the user asks for new content \u2014 a\ndesign, a deck, a form, a document \u2014 author it yourself and save it with the\napp's create/update tools. Never hand off authoring to the app's own agent\nwhen those tools exist, and never wait on an in-app question form; it answers\nback through that app's own chat, not to you.\n\nUse `ask_app` only when the host has no direct page tools, the requested\ncapability is not exposed, the task needs the app agent's interpretation or\nmulti-step specialist reasoning, or a direct call fails and needs recovery.\nDispatch's unified `/mcp` endpoint still exposes generic cross-app verbs; a\ndirect app MCP connection or page WebMCP surface is where named app actions\nappear.\n\nThe app's MCP `instructions` name its key tools (generated from the app's own\nlist); follow those names verbatim and use `tool-search` for anything else \u2014\nnever guess a tool name.\n\n1. Call `view-screen` before any selection-dependent edit and use the ids it\n   returns; never guess an id from a title, index, or stale context.\n2. If `view-screen` returns a `selection` and `nextRequiredAction`, make\n   that call directly with the smallest atomic change; do not regenerate a\n   whole artifact for a focused edit.\n3. Read the result back with the app's matching read action and report what\n   changed. A queued `ask_app_status` task is not proof an edit completed.\n\n## Authentication and host behavior\n\nThe Dispatch MCP connector identifies the calling agent, while the embedded app\nuses the user's browser session. Let the user complete sign-in in the rendered\nAgent-Native browser surface. Never request, copy, or store passwords, OAuth\ncodes, cookies, or access tokens.\n\nThe generic Agent-Native plugin registers Dispatch at\n`https://dispatch.agent-native.com/mcp` for hosts that support plugin MCP\nconfiguration. ChatGPT custom connectors use the same URL with OAuth. After\ninstalling or changing a connector, reload Claude/Cowork/ChatGPT or reconnect\nand rescan the MCP server if the new tools or inline surface are not visible.\n\n## Failure handling\n\n- If `open_app` succeeds but no inline surface appears, show its returned\n  link and say that this host does not expose an embedded browser here.\n- After an app loads, check for `list-host-webmcp-tools` before delegating. If\n  it is available, use the exact returned tool name and origin for direct work.\n- If `ask_app` cannot reach Slides, check that Slides is granted in\n  Dispatch and that the connector is authenticated, then retry once with the\n  exact app id `slides`.\n- If screen state has no deck or selection, describe that state instead of\n  fabricating context. Ask the user to open a deck or select an element when\n  the requested edit depends on it.\n- Treat a successful open link, a queued task, or a tool acknowledgment as\n  navigation/task setup only. Confirm a write with the app's readback action.\n";
export declare const AN_COMMAND_MD = "---\ndescription: Open and operate a granted Agent-Native app through Dispatch MCP.\nargument-hint: <app> [request]\n---\n\nUse the an skill. Treat the first argument as the Agent-Native Dispatch app id.\nFor `/an slides`, call `open_app` with app `slides`, path \"/\", embed\ntrue, and chrome `full`. Keep the returned link if the host cannot render\nthe inline MCP App. Any remaining arguments are the user's request after the\napp opens:\n\n$ARGUMENTS\n";