/rubber-duck fails with GPT-6 Astra: no complementary model available (Windows, app 1.1.20)

Riproduci /rubber-duck review con GPT-6 Astra nell’app GitHub Copilot, quindi traccia il task rubber-duck e il percorso di selezione del modello di /subagents usando l’errore segnalato e il registro degli eventi. Confronta il comportamento con un modello di subagent esplicito e con le segnalazioni #3791 e #3491; il lavoro è completato quando la review si avvia con un modello complementare oppure l’app mostra un errore di preflight chiaro e azionabile.

Summary

/rubber-duck review fails to start in a GPT-6 Astra session. The app accepts the command, but the Rubber Duck task immediately fails because no complementary model is available for the current session model.

Environment

Field	Value
GitHub Copilot app version	1.1.20 (recorded by the app in the original report)
Operating system	Windows 11 Enterprise, version 25H2, OS build 26200.9445, 64-bit
Session model	GPT-6 Astra (gpt-6-astra), confirmed in the failed task event
Affected workflow	Review in an existing local project session
Failure observed	September 14, 2026, 10:41:35 UTC+08:00

Steps to reproduce

1. Open an existing local project session in the GitHub Copilot app.
2. Select GPT-6 Astra as the session model.
3. After working on changes, invoke /rubber-duck review without specifying an explicit subagent model.
4. Observe the Rubber Duck task fail immediately instead of starting a review.

Expected behavior

Rubber Duck should start an independent review using an available complementary model.

If the current session/model configuration is unsupported, the app should indicate this before starting the task and provide an actionable model-selection path, rather than accepting the command and failing immediately.

Actual behavior

• The Rubber Duck task displays a failure indicator and "Worked for less than a second."
• The session event log records a failed task invocation with agent_type: "rubber-duck". The start and failure timestamps are approximately 60 ms apart.
• The runtime reports:

The "rubber-duck" subagent uses a complementary model (a different model family than your session), but none is available for your current session model. Pick a specific model for it in /subagents, or switch your session model.

• The assistant asks the user to select a model. In the captured run, the user was unavailable, so it continued with separate General Purpose and Code Review agents and explicitly stated that this was not a cross-model Rubber Duck review.

Screenshot evidence

The reporter's screenshot shows the failed Rubber Duck task and a model-selection question. English translation of the displayed question:

Rubber-duck review could not start: the current model has no complementary model available for automatic selection. Which model should be specified for this independent review?

The screenshot has been cropped locally to exclude unrelated project information. The image attachment is pending.

Diagnostic log excerpt

The following is a minimal, sanitized excerpt from the affected session's event log, not a /collect-debug-logs archive. Private review prompts, local paths, and session/call identifiers have been omitted.

2026-09-14T02:41:35.514Z tool.execution_start
  toolName: task
  arguments.agent_type: rubber-duck
  arguments.mode: sync

2026-09-14T02:41:35.574Z tool.execution_complete
  model: gpt-6-astra
  success: false
  error.code: failure
  error.message: [exact message quoted above]

A /collect-debug-logs archive has not been collected or attached for this update.

Additional context

The log confirms the failed model selection, but does not establish why no complementary model was available. I have not verified whether explicitly selecting a model in /subagents or changing the session model resolves the problem.

Potentially related reports: #3791 and #3491. This report adds a Windows 1.1.20 occurrence with the exact runtime error.