fix: recompute Copilot api_mode after model switch

Recomputes GitHub Copilot api_mode from the selected model in the
shared /model switch path.  Before this change, Copilot could carry a
stale codex_responses mode forward from a GPT-5 selection into a later
Claude model switch, causing unsupported_api_for_model errors.

Cherry-picked from #10533 by @helix4u with:
- Comment specificity (Provider-specific → Copilot api_mode override)
- Fix pre-existing duplicate opencode-go in set literal
- Extract test mock helper to reduce duplication
- Add GPT-5 → GPT-5 regression test (keeps codex_responses)

hermes_cli/model_switch.py

@@ -457,6 +457,7 @@ def switch_model(
         ModelSwitchResult with all information the caller needs.
     """
     from hermes_cli.models import (
+        copilot_model_api_mode,
         detect_provider_for_model,
         validate_requested_model,
         opencode_model_api_mode,
@@ -714,8 +715,12 @@ def switch_model(
     if validation.get("corrected_model"):
         new_model = validation["corrected_model"]
 
+    # --- Copilot api_mode override ---
+    if target_provider in {"copilot", "github-copilot"}:
+        api_mode = copilot_model_api_mode(new_model, api_key=api_key)
+
     # --- OpenCode api_mode override ---
-    if target_provider in {"opencode-zen", "opencode-go", "opencode", "opencode-go"}:
+    if target_provider in {"opencode-zen", "opencode-go", "opencode"}:
         api_mode = opencode_model_api_mode(target_provider, new_model)
 
     # --- Determine api_mode if not already set ---
@@ -1098,5 +1103,3 @@ def list_authenticated_providers(
     results.sort(key=lambda r: (not r["is_current"], -r["total_models"]))
 
     return results
-
-

tests/hermes_cli/test_model_switch_copilot_api_mode.py

@@ -0,0 +1,101 @@
+"""Regression tests for Copilot api_mode recomputation during /model switch.
+
+When switching models within the Copilot provider (e.g. GPT-5 → Claude),
+the stale api_mode from resolve_runtime_provider must be overridden with
+a fresh value computed from the *new* model.  Without the fix, Claude
+requests went through the Responses API and failed with
+``unsupported_api_for_model``.
+"""
+
+from unittest.mock import patch
+
+from hermes_cli.model_switch import switch_model
+
+
+_MOCK_VALIDATION = {
+    "accepted": True,
+    "persist": True,
+    "recognized": True,
+    "message": None,
+}
+
+
+def _run_copilot_switch(
+    raw_input: str,
+    current_provider: str = "copilot",
+    current_model: str = "gpt-5.4",
+    explicit_provider: str = "",
+    runtime_api_mode: str = "codex_responses",
+):
+    """Run switch_model with Copilot mocks and return the result."""
+    with (
+        patch("hermes_cli.model_switch.resolve_alias", return_value=None),
+        patch("hermes_cli.model_switch.list_provider_models", return_value=[]),
+        patch(
+            "hermes_cli.runtime_provider.resolve_runtime_provider",
+            return_value={
+                "api_key": "ghu_test_token",
+                "base_url": "https://api.githubcopilot.com",
+                "api_mode": runtime_api_mode,
+            },
+        ),
+        patch(
+            "hermes_cli.models.validate_requested_model",
+            return_value=_MOCK_VALIDATION,
+        ),
+        patch("hermes_cli.model_switch.get_model_info", return_value=None),
+        patch("hermes_cli.model_switch.get_model_capabilities", return_value=None),
+        patch("hermes_cli.models.detect_provider_for_model", return_value=None),
+    ):
+        return switch_model(
+            raw_input=raw_input,
+            current_provider=current_provider,
+            current_model=current_model,
+            explicit_provider=explicit_provider,
+        )
+
+
+def test_same_provider_copilot_switch_recomputes_api_mode():
+    """GPT-5 → Claude on copilot: api_mode must flip to chat_completions."""
+    result = _run_copilot_switch(
+        raw_input="claude-opus-4.6",
+        current_provider="copilot",
+        current_model="gpt-5.4",
+    )
+
+    assert result.success, f"switch_model failed: {result.error_message}"
+    assert result.new_model == "claude-opus-4.6"
+    assert result.target_provider == "copilot"
+    assert result.api_mode == "chat_completions"
+
+
+def test_explicit_copilot_switch_uses_selected_model_api_mode():
+    """Cross-provider switch to copilot: api_mode from new model, not stale runtime."""
+    result = _run_copilot_switch(
+        raw_input="claude-opus-4.6",
+        current_provider="openrouter",
+        current_model="anthropic/claude-sonnet-4.6",
+        explicit_provider="copilot",
+    )
+
+    assert result.success, f"switch_model failed: {result.error_message}"
+    assert result.new_model == "claude-opus-4.6"
+    assert result.target_provider == "github-copilot"
+    assert result.api_mode == "chat_completions"
+
+
+def test_copilot_gpt5_keeps_codex_responses():
+    """GPT-5 → GPT-5 on copilot: api_mode must stay codex_responses."""
+    result = _run_copilot_switch(
+        raw_input="gpt-5.4-mini",
+        current_provider="copilot",
+        current_model="gpt-5.4",
+        runtime_api_mode="codex_responses",
+    )
+
+    assert result.success, f"switch_model failed: {result.error_message}"
+    assert result.new_model == "gpt-5.4-mini"
+    assert result.target_provider == "copilot"
+    # gpt-5.4-mini is a GPT-5 variant — should use codex_responses
+    # (gpt-5-mini is the special case that uses chat_completions)
+    assert result.api_mode == "codex_responses"