fix(chat): resolve session model before activating

static/sessions.js

@@ -567,10 +567,14 @@ async function loadSession(sid){
     if (_msgInner && currentSid !== sid) _msgInner.innerHTML = '<div style="display:flex;align-items:center;justify-content:center;height:100%;color:var(--text-muted);font-size:14px;padding:40px;text-align:center;">Loading conversation...</div>';
   }
   // Phase 1: Load metadata only (~1KB) for fast session switching.
+  // Resolve model immediately: old sessions can persist stale provider-shaped
+  // IDs (e.g. openai/gpt-5.4-mini) and assigning those to S.session creates a
+  // short race where the composer can display/send the wrong model before the
+  // deferred resolver catches up.
   // Guard against network/server failures to prevent a permanently stuck loading state.
   let data;
   try {
-    data = await api(`/api/session?session_id=${encodeURIComponent(sid)}&messages=0&resolve_model=0`);
+    data = await api(`/api/session?session_id=${encodeURIComponent(sid)}&messages=0&resolve_model=1`);
   } catch(e) {
     const _msgInner = $('msgInner');
     if(_msgInner){

tests/test_session_model_resolution_on_load.py

@@ -0,0 +1,46 @@
+"""Regression tests for stale session model hydration in the WebUI.
+
+Old sessions can persist provider-shaped model IDs such as ``openai/gpt-5.4-mini``
+after the active runtime moved to OpenAI Codex ``gpt-5.5``.  The first
+``loadSession()`` metadata request must ask the backend for the resolved model so
+that the composer state cannot briefly use the stale raw value for display or the
+next chat-start payload.
+"""
+
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+SESSIONS_JS = (REPO_ROOT / "static" / "sessions.js").read_text(encoding="utf-8")
+
+
+def _extract_function(src: str, signature: str) -> str:
+    start = src.find(signature)
+    assert start >= 0, f"missing function signature: {signature}"
+    brace = src.find("{", start)
+    assert brace >= 0, f"missing function body for: {signature}"
+    depth = 0
+    for idx in range(brace, len(src)):
+        ch = src[idx]
+        if ch == "{":
+            depth += 1
+        elif ch == "}":
+            depth -= 1
+            if depth == 0:
+                return src[start : idx + 1]
+    raise AssertionError(f"unterminated function body for: {signature}")
+
+
+def test_load_session_initial_metadata_request_resolves_model_before_state_assignment():
+    body = _extract_function(SESSIONS_JS, "async function loadSession(sid")
+    metadata_fetch = "messages=0&resolve_model=1"
+    stale_metadata_fetch = "messages=0&resolve_model=0"
+    assignment = "S.session=data.session"
+
+    assert metadata_fetch in body, (
+        "loadSession() must resolve model metadata on the initial fetch so stale "
+        "persisted models like openai/gpt-5.4-mini cannot become active composer state"
+    )
+    assert stale_metadata_fetch not in body[: body.index(assignment)], (
+        "loadSession() must not assign S.session from unresolved metadata before the "
+        "backend has normalized stale model/provider combinations"
+    )