Executor

What Executor is

Executor is a host that runs apps: TypeScript packages that expose tools, storage, UI, accounts and background work. Agents reach every app through the MCP tools skills, search, describe, call and execute, with each app's slug as its namespace. Credentials go only through Executor's secure connection links, never through chat or source code. Some calls pause for user approval.

This is the Executor app's executor skill. Read its other skills with skills({app: "executor", name: "code-mode"}); skills({}) lists them all.

Find and call tools

search({ query: "..." }) returns each tool's exact path, such as tools.<slug>.<tool>, and input type; describe({ paths }) returns full signatures. Never guess argument names.

For a few independent calls, use call({ calls: { path, input } }); an array of calls runs concurrently. Each result has ok and its value or error.

Write a JavaScript program for execute when one call's result feeds another, or a large result should be filtered before it reaches you. Inside it, the same paths are async functions, and tools.search and tools.search.describe work as above. Return only the data you need. For depth, read the code-mode skill. When call or execute pauses for approval or input, show the request to the user and resume it; never repeat the call or run the program again.

Build an app

Build an app when no existing app provides the tools, data, UI or automation the user needs. For a known service, search the catalog first: organization.catalog({ query: { search: "gmail", limit: "20" } }), then apps.install a match or apps.importCustom a remote MCP URL. With a shell, use the executor apps CLI, else execute; read the app-authoring skill first.

Send feedback

Send feedback with the Executor app's feedback.submit tool when Executor gets in your way, can't do something you need, works especially well, or the user has feedback. Before your final reply, check whether anything is worth reporting. Be specific; never include credentials or user data (feedback.md).