openclaw - 💡(How to fix) Fix [Bug]: session_status tool reports stale model after runtime fallback — shows configured model instead of actual active model [1 comments, 2 participants]

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

The session_status tool reads entry.model from the session store, which is only written on explicit user-driven model changes (/model or sessions.patch). When a runtime fallback occurs (e.g., primary model hits rate limit), the ACP runtime switches to a fallback model, but entry.model is not updated. This causes session_status to report the stale configured model while the UI status bar (which reads actual runtime state) shows the correct fallback model.

This discrepancy was observed when the primary model deepseek/deepseek-v4-flash fell back to minimax-portal/MiniMax-M2.7-highspeed — the session_status tool still reported deepseek/deepseek-v4-flash.

Expected behavior

sesson_status should report the actual active model that is currently handling requests in the session, not just the configured/default model.

Steps to reproduce

1. Configure a primary model with fallback models (e.g., DeepSeek V4 Flash with MiniMax-M2.7-highspeed as fallback)
2. Trigger a condition that causes fallback (rate limit, error, etc.)
3. Call the session_status tool
4. Observe: tool reports configured primary model, not the actual fallback model
5. Compare with UI status bar — they will disagree

Environment

• OpenClaw: 2026.4.27 (cbc2ba0)
• Channel: Feishu webchat
• Runtime: ACP with model fallback chain

Technical root cause

In session-status-tool.ts → createSessionStatusTool → execute:

const runtimeModelIdentity = resolveSessionModelIdentityRef(cfg, resolved.entry, agentId, \`\${configured.provider}/\${configured.model}\`);

resolveSessionModelIdentityRef() only reads entry.model and entry.modelProvider. These fields are only written on explicit model changes, not on runtime fallback events.

The actual runtime model (including fallback state) is tracked separately in the ACP session runtime, but sesson_status does not read it.

Relevant code paths:

• resolveSessionModelIdentityRef() in session-utils-DwIoA6Si.js — reads entry.model / entry.modelProvider
• resolveSelectedAndActiveModel() in model-runtime-DATWIRIw.js — correctly distinguishes selected vs active, but sesson_status only uses the selected/configured model for the status card
• resolvePersistedModelRef() in model-selection-DGI1ibK2.js — only checks stored overrides and entry.model, not runtime fallback state

Suggested fixes

1. Best: Have session_status read the actual active runtime model from the ACP session runtime (same source the UI status bar uses)
2. Alternative: Add a note/warning in the status card when selected !== active to indicate a fallback is in effect
3. Alternative: Update entry.model when runtime fallback occurs (requires ACP runtime to write back to session store)

Related

This is related to how the model fallback mechanism works — the fallback is transparent to the agent at the tool level, which is by design for resilience, but sesson_status should ideally surface this information accurately.