Model Selection

Model Selection

Use this skill when the user requests a model, the team configuration contains a saved model preference, or a specialist charter explicitly requires one.

Resolution Order

Apply the first available value:

1. Explicit model requested for the current spawn.
2. Current-session model directive from the user.
3. .squad/config.json agentModelOverrides.{specialist-id}.
4. .squad/config.json defaultModel.
5. The specialist charter's explicit model requirement.
6. Runtime automatic selection by omitting the model parameter.

An explicit current request always overrides older persistent configuration. Do not invent a model preference from task type alone, and do not pin a hardcoded fallback that can become obsolete.

Persistent Preferences

• Save defaultModel or agentModelOverrides only when the user explicitly asks for persistence.
• Merge model fields into the existing .squad/config.json; do not replace unrelated settings.
• Use active specialist IDs from .squad/casting/registry.json.
• Validate a requested model against the runtime's currently supported model list.
• To return to automatic routing, remove the applicable override rather than writing an "auto" sentinel unless the runtime schema explicitly supports it.

Spawn Behavior

• Pass a model only when resolution steps 1-5 produce a supported value.
• Otherwise omit the model parameter and allow the runtime to select.
• Preserve the user's requested reasoning effort or context tier only when the runtime supports it.
• If a requested model is unavailable, report that failure and fall back to runtime automatic selection unless the user supplied a different fallback.

Stop Conditions

• Stop if the requested model is not supported and no automatic fallback is acceptable.
• Stop rather than rewriting configuration when .squad/config.json cannot be parsed safely.
• Do not benchmark models, estimate billing, or modify specialist responsibilities under this skill.