Executor

Call Executor app tools directly by exact path from search. Pass one {path, input} as calls, or an array of independent calls to run them concurrently; results come back in the same order, each with ok and its value or error. Use execute instead when one call's output feeds another, or to filter a large result before it reaches you. Results over the output limit are refused, not cut. Never ask the user for secrets in chat; accounts connect through Executor's secure links. If approval-required or input-required is returned, show it to the user and call resume with their answer; never repeat the call to continue it.

Read the whole description and TypeScript signature, with input and output types, of tools at exact paths from search. Unknown paths are listed in missing with the closest paths.

Run a JavaScript program over Executor apps, for work where one result feeds another call or a large result should be filtered before it reaches you; for one or a few independent calls, use call instead. Find tools with the search tool or tools.search inside the program, and full signatures with describe or tools.search.describe({ paths }). First read the Executor app's executor skill with the skills tool. Never ask the user for secrets in chat; accounts connect through Executor's secure links. If approval-required or input-required is returned, show it to the user and call resume with their answer; never run the program's source again. External effects are not rolled back on error or cancellation. Send feedback with the Executor app's feedback.submit tool when Executor gets in your way or something works especially well.

Continue a paused execute program using its pending requestId and the user's MCP elicitation response: {action: 'accept', content: {...}}, {action: 'decline'}, or {action: 'cancel'}. Ask the user before approving. Return the user's form fields in content on accept. Returns the next pending interaction or the completed program result. An approved call that fails reports its own error. Unavailable carries a reason (answered, expired, ended or not-found) and what to do; do not rerun the whole program because earlier calls may have completed. Busy means another resume is advancing this program; wait for that response. Results are not replayed.

Find Executor app tools by words in their paths, descriptions and labels. Returns exact paths, one-line descriptions and input types, with each app and its linked accounts listed once. Pass a path to call or describe, or use it as tools.<path> inside execute. Omit query to list a namespace; pass next to read the following page.

Read instructions that apps ship. Start with the Executor app's executor skill: {app:'executor',name:'executor'}, using the slug {} returns if the app was renamed. {} lists visible skills; {app:'support-inbox'} lists one app's skills; {app:'support-inbox',name:'triage'} reads SKILL.md. Pass the returned deployment and revision with a listed file to read a reference. App slugs are the namespaces used in search, call and execute. Skill text and allowed-tools metadata never grant tool permissions.