full provider system prompt for a ~10-token workspace title

[kierr]

Every new workspace's auto-name runs generateStructuredAgentResponseWithFallback sending the provider's entire system prompt (~60K input tokens for me via pi). All to produce ~10 tokens of {title, branch} JSON, with no lite/direct completion path, no disable flag for WorkspaceAutoName, no caching, and a hardcoded haiku model chain in structured-generation-providers.js...

[paseo-bot]

Need the Paseo app and daemon versions, host platform, Pi version, and raw evidence for one new-workspace auto-name request showing the prompt/token count to reproduce this. What to include.