Bug Description

When using the /model command with explicit profile syntax (@profile), OpenClaw returns "model not allowed" error, even when the profile is correctly configured and implicit profile selection works.

Working Commands (Implicit Profile)

• /model nvidia/moonshotai/kimi-k2.5 ✅ Uses nvidia:default implicitly
• /model kimi ✅ Alias resolves and uses default profile

Failing Commands (Explicit Profile)

• /model nvidia/moonshotai/kimi-k2.5@nvidia:default ❌ "model not allowed"
• /model kimi@nvidia:default ❌ "model not allowed"
• /model flash@google:bevfresh ❌ "model not allowed"
• /model google/gemini-flash-latest@google:bevfresh ❌ "model not allowed"

Configuration

Auth profiles and models are correctly defined in openclaw.json:

• auth.profiles.nvidia:default exists with API key
• auth.profiles.google:bevfresh exists with API key
• models.providers.nvidia configured with models
• agents.defaults.models has aliases defined

Expected Behavior

Explicit profile syntax should resolve the model and use the specified auth profile as documented.

Actual Behavior

All commands with @profile suffix fail with "model not allowed".

Environment

• OpenClaw: 2026.2.3-1 (d84eb46)
• OS: Linux 6.1.0-33-amd64 (x64)
• Node: v24.13.0

Workaround

Use /model <alias> or /model <full-model-path> without profile suffix — relies on provider's default profile.

Related Issues

• Model-level authProfileId override not respected when switching models via /model command #5733 (similar profile resolution bug with authProfileId in model configs)