Author: philz42
Channel: #shelley
Link: https://discord.com/channels/1405685085923049482/1450334528210993295/1481804995182592033

Issue

The subagent tool has the capacity to choose which model to use, but it's either explicitly or implicitly using the wrong one. This causes issues when users have configured API keys for specific models, as the agent falls back to incorrect models instead of respecting the parent agent's model choice.

Requested Changes

1. Short-term fix: Update agents.md documentation to instruct subagents to use the same model as the parent agent consistently
2. Long-term improvement: Implement the ability to disable built-in models so users can enforce use of their configured models

Context

User reported that when spawning sub-agents, the model selection doesn't match the parent agent's model. In their case, it was falling back to exe.dev's Claude (with no credit) instead of their own API key (which has available credit). This may be related to recent changes in how Shelley spawns sub-agents and VM region migrations.

Related change: Full context was recently removed from sub-agents to improve load times; now only patches are passed instead of old & new versions.