ACP mode: session/set_model rejects "auto" but rate-limit error suggests switching to it

[catatafishen]

Describe the bug

In ACP mode (copilot --acp --stdio), when a model-specific rate limit is hit, the CLI surfaces this message to the client:

You've reached your weekly rate limit. Please wait for your limit to reset in N hours or switch to auto model to continue.

However, sending session/set_model with modelId: "auto" immediately returns an error:

{
  "code": -32602,
  "message": "Invalid model 'auto'. Supported values: claude-sonnet-4.6, claude-haiku-4.5, claude-opus-4.6, gpt-5.4, gpt-5.3-codex, gpt-5-mini, gpt-4.1."
}

So the rate-limit message instructs the user (and ACP clients) to switch to the auto model, but the ACP protocol itself rejects the "auto" model ID. There is no ACP method that exposes auto-model selection at all.

Reproduction (raw ACP JSON-RPC)

import subprocess, json, os

copilot = "<path-to-copilot-binary>"
proc = subprocess.Popen([copilot, "--acp", "--stdio", "--disable-builtin-mcps", "--no-auto-update"],
    stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, bufsize=1)

def send(method, params=None, id=None):
    msg = {"jsonrpc": "2.0", "method": method}
    if params: msg["params"] = params
    if id is not None: msg["id"] = id
    proc.stdin.write(json.dumps(msg) + "\n"); proc.stdin.flush()

def recv(): return json.loads(proc.stdout.readline())

send("initialize", {"protocolVersion": 1, "clientInfo": {"name":"Test","title":"Test","version":"1"}, "clientCapabilities": {}}, id=1)
recv()
send("initialized", {})
send("session/new", {"cwd": "/tmp", "mcpServers": []}, id=2)
r = recv()
session_id = r["result"]["sessionId"]

send("session/set_model", {"sessionId": session_id, "modelId": "auto"}, id=3)
print(recv())
# {"jsonrpc": "2.0", "id": 3, "error": {"code": -32602, "message": "Invalid model 'auto'..."}}

Expected behavior

Either:

1. session/set_model accepts "auto" as a valid modelId in ACP mode (matching its availability in the web UI and GitHub Actions), or
2. The rate-limit error message does not suggest switching to auto model when the client is in ACP mode where it cannot be done.

Affected version

GitHub Copilot CLI 1.0.32

Additional context

• "auto" is accepted as a model ID in Copilot Cloud / GitHub Actions contexts (see Request to Copilot Cloud Agent via custom agent fails with 400 error if model selection does not align #2695), so the backend clearly supports it — it just is not exposed via ACP's session/set_model.
• ACP clients that display the rate-limit message to users (IDE plugins, third-party tools) have no way to act on the "switch to auto model" suggestion.
• Related: Weekly rate limiting #2828 (rate-limit message UX), Rate limit aborts subagents #2840 (auto model issues with sub-agents)

[devm33]

Hi @catatafishen auto mode will be available via ACP in the next release