Problem

When overriding Sisyphus to use a GPT model (e.g., github-copilot/gpt-5.2), the agent retains Claude-specific thinking config instead of applying GPT-appropriate options like reasoningEffort.

This matters for GitHub Copilot users where Anthropic Opus consumes significantly more premium requests, making GPT-5.2 the practical orchestrator model.

Solution

Auto-detect the model provider and apply appropriate options in src/agents/utils.ts:

For GPT models (openai/*, github-copilot/gpt-*):

• Apply reasoningEffort, textVerbosity defaults
• Remove Claude-specific thinking config

For Anthropic models (anthropic/*):

• Apply thinking config with budgetTokens

Example

{
  "agents": {
    "Sisyphus": {
      "model": "github-copilot/gpt-5.2"
    }
  }
}

System detects GPT model and applies appropriate reasoning options automatically.

Dependency

For GitHub Copilot, reasoningEffort requires GPT-5+ models routed through the Responses API: sst/opencode#5877