dsh-codex-reasoning-router

Configuration

Host configuration lives in cordis.patch.yml:

- id: reasoning-router
  name: dsh-codex-reasoning-router
  inject: [openAICodex]
  config:
    requiredPresetId: luna-sol-reasoning-router
    lunaProvider: openai-codex
    lunaModel: gpt-5.6-luna
    solProvider: openai-codex
    solModel: gpt-5.6-sol
    initialSolReasoning: medium
    escalatedSolReasoning: high
    solAdviceMaxTokens: 2000
    solTimeoutMs: 30000
    initialConsultEnabled: true
    failOpen: true

Both effort fields accept only medium or high and default to medium -> high. The narrow domain type and runtime guard reject xhigh and max. When a root using this preset starts, the plugin checks the provider catalog and exact model metadata; a missing configured model is an error, not a fallback. That check is not performed for other presets.

Confirm current IDs with the DSH model picker (/model) or the public LLM registry used by a diagnostic plugin:

await ctx.llm.listModels('openai-codex')
await ctx.llm.resolveModelInfo('openai-codex', 'gpt-5.6-luna')
await ctx.llm.resolveModelInfo('openai-codex', 'gpt-5.6-sol')

At implementation time, the installed pi-ai catalog contains gpt-5.6-luna and gpt-5.6-sol.