@Developer Tools Engineer
Thank you. I’m building a custom ACP-based coding agent that I’m currently calling TaylorACP. It is not a publicly released agent; it is a project I’m developing to connect Xcode to hosted models through OpenRouter.
The agent already supports Xcode’s normal ACP session flow, plan mode, transcript streaming, tool calls, filesystem access, source edits, diff artifacts, and permission requests.
For plans, I have tested:
- ACP v1
session/update with sessionUpdate: "plan" and structured entries - ACP v2
session/update with sessionUpdate: "plan_update" and an item-based plan plan_update containing the completed Markdown plan- A
switch_mode tool call with session/request_permission
The plan text appears in the conversation transcript, and some attempted payloads produce a Markdown/card-style item. However, the completed plan does not appear as the native editable Markdown artifact in Xcode’s right-side artifact pane.
The switch_mode flow correctly produces the “Ready to code?” approval prompt, but it does not first create the editable plan artifact. Edit or diff tool calls create normal source-file artifacts rather than a plan artifact.
Because this is a custom agent I am building, I can modify its ACP output to emit whatever signal Xcode requires. I can also provide a minimal reproduction or protocol trace showing the exact messages it currently sends.
What exact ACP signal or message sequence does Xcode expect from a custom agent to indicate that a plan is complete and should appear as an editable Markdown artifact in the artifact pane?
