skills.rest
seiggy
seiggyCOMMUNITY
Explore all skills by this author →
GitHub

model-selection

Resolves the LLM model for each agent spawn using a five-layer priority hierarchy.

1|Updated Jul 7, 2026
Search Tags:#agent orchestration#llm routing#model selection#config management#claude models#fallback chains#task-aware selection
One-click install
npx skills add https://github.com/seiggy/maf-copilot-studio-demo --skill model-selection-seiggy
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: model-selection
Source: https://github.com/seiggy/maf-copilot-studio-demo/tree/main/.squad/templates/skills/model-selection
Command: npx skills add https://github.com/seiggy/maf-copilot-studio-demo --skill model-selection-seiggy

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinators spawning multiple agents need a consistent, predictable way to pick the right LLM model for each task without manually specifying it every time, while still honoring persistent user preferences. ## Core Features & Use Cases - Five-Layer Resolution Hierarchy: Resolves models in priority order from persistent config overrides, session directives, agent charter preferences, task-aware auto-selection, down to a hardcoded default. - Persistent Preferences: Saves defaultModel and per-agent agentModelOverrides to .squad/config.json so choices survive across sessions. - Task-Aware Auto-Selection: Maps task types to tiers, sending code work to claude-sonnet-4.6, visual analysis to claude-opus-4.6, and docs or triage to claude-haiku-4.5. - Tier-Constrained Fallback Chains: Retries within the same tier when a model is unavailable, never falling up from fast to premium. - Use Case: A user says "always use opus for Fenster"; the skill validates the model ID, writes the override to config.json, and every future spawn of that agent acknowledges the resolved model. ## Quick Start Ask the coordinator to set a default model, for example by saying "always use claude-sonnet-4.6 for everything", and it will save the preference and apply it to all future agent spawns.

Frequently Asked Questions about model-selection

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I set a default model for all agent spawns?▼

Say a trigger phrase like "always use X" or "default to X" and the skill validates the model ID against the catalog, then writes defaultModel to .squad/config.json. The preference persists across sessions until cleared.

How do I assign a specific model to one agent?▼

Use a per-agent trigger such as "use opus for Fenster". The skill writes the model to agentModelOverrides under that agent's name in .squad/config.json, which takes priority over the global defaultModel.

Which model is selected for coding tasks automatically?▼

Code tasks including implementation, tests, refactoring, and bug fixes resolve to claude-sonnet-4.6 at the task-aware layer. Non-code work like docs and triage maps to claude-haiku-4.5, and visual analysis maps to claude-opus-4.6.

What happens if the selected model is unavailable?▼

The skill retries within the same tier using defined fallback chains, for example claude-opus-4.6 falls back to claude-opus-4.6-fast then claude-opus-4.5. It never falls up to a higher tier, so fast tasks never land on premium models.

How do I clear a saved model preference?▼

Say "switch back to automatic", "clear model preference", or "use default models". The skill removes defaultModel from .squad/config.json and resolution returns to the automatic layers.

Does a session directive override a saved config preference?▼

No. Layer 0 persistent config in .squad/config.json beats everything, including session directives and charter preferences. A saved defaultModel applies to all spawns until explicitly cleared.

Related Skills

Explore More Software Engineering Skills→
DeDuvaDeDuva
1

model-selection

Assign LLM models to agent spawns using a multi-layer resolution hierarchy.

Community
Advanced
quaid-appquaid-app
3

Model Selection

Resolve and persist LLM model choices per agent spawn in .squad/config.json.

Official
Advanced
ruvnetruvnet
70.1k

cost-track

Captures per-session token usage from Claude Code session jsonl and persists cost records to memory.

Community
Intermediate
xzedmxzedm

review-loop

Iterates work through a critic subagent scoring cycle until a quality gate is met.

Community
Intermediate