model-selection

name	model-selection
description	Per-agent model selection with 4-layer hierarchy and fallback chains
domain	orchestration
confidence	high
source	extracted

Context

Before spawning an agent, the coordinator determines which model to use. This skill codifies the 4-layer hierarchy, role-to-model mappings, task complexity adjustments, and fallback chains. Applies to all agent spawns in Team Mode.

Patterns

4-Layer Hierarchy

Check these layers in order — first match wins:

Layer 1 — User Override: Did the user specify a model? ("use opus", "save costs", "use gpt-5.3-codex for this"). If yes, use that model. Session-wide directives ("always use haiku") persist until contradicted.

Layer 2 — Charter Preference: Does the agent's charter have a ## Model section with Preferred set to a specific model (not auto)? If yes, use that model.

Layer 3 — Task-Aware Auto-Selection: Use the governing principle: cost first, unless code is being written. Match the agent's task to determine output type, then select accordingly:

Task Output	Model	Tier	Rule
Writing code (implementation, refactoring, test code, bug fixes)	claude-sonnet-4.6	Standard	Quality and accuracy matter for code. Use standard tier.
Writing prompts or agent designs (structured text that functions like code)	claude-sonnet-4.6	Standard	Prompts are executable — treat like code.
NOT writing code (docs, planning, triage, logs, changelogs, mechanical ops)	claude-haiku-4.5	Fast	Cost first. Haiku handles non-code tasks.
Visual/design work requiring image analysis	claude-opus-4.8	Premium	Vision capability required. Overrides cost rule.

Role-to-model mapping (applying cost-first principle):

Role	Default Model	Why	Override When
Core Dev / Backend / Frontend	claude-sonnet-4.6	Writes code — quality first	Heavy code gen → gpt-5.3-codex
Tester / QA	claude-sonnet-4.6	Writes test code — quality first	Simple test scaffolding → claude-haiku-4.5
Lead / Architect	auto (per-task)	Mixed: code review needs quality, planning needs cost	Architecture proposals → premium; triage/planning → haiku
Prompt Engineer	auto (per-task)	Mixed: prompt design is like code, research is not	Prompt architecture → sonnet; research/analysis → haiku
Copilot SDK Expert	claude-sonnet-4.6	Technical analysis that often touches code	Pure research → claude-haiku-4.5
Designer / Visual	claude-opus-4.8	Vision-capable model required	— (never downgrade — vision is non-negotiable)
DevRel / Writer	claude-haiku-4.5	Docs and writing — not code	—
Scribe / Logger	claude-haiku-4.5	Mechanical file ops — cheapest possible	— (never bump Scribe)
Git / Release	claude-haiku-4.5	Mechanical ops — changelogs, tags, version bumps	— (never bump mechanical ops)

Task complexity adjustments (apply at most ONE — no cascading):

• Bump UP to premium: architecture proposals, reviewer gates, security audits, multi-agent coordination (output feeds 3+ agents)
• Bump DOWN to fast/cheap: typo fixes, renames, boilerplate, scaffolding, changelogs, version bumps
• Switch to code specialist (gpt-5.3-codex): large multi-file refactors, complex implementation from spec, heavy code generation (500+ lines)
• Switch to analytical diversity (gemini-2.5-pro): code reviews where a second perspective helps, security reviews, architecture reviews after a rejection

Layer 4 — Default: If nothing else matched, use claude-haiku-4.5. Cost wins when in doubt, unless code is being produced.

Fallback Chains

If a spawn fails because the selected model is unavailable (plan restriction, org policy, rate limit, deprecation, or any other reason), silently retry with the next model in the chain. Do NOT tell the user about fallback attempts. Maximum 3 retries before jumping to the nuclear fallback.

Premium:  claude-opus-4.8 → claude-opus-4.7 → claude-opus-4.6 → claude-sonnet-4.6 → (omit model param)
Standard: claude-sonnet-5 → claude-sonnet-4.6 → gpt-5.6-sol → gpt-5.6-terra → gpt-5.6-luna → gpt-5.4 → gpt-5.3-codex → claude-sonnet-4.5 → gemini-2.5-pro → (omit model param)
Fast:     claude-haiku-4.5 → gpt-5.4-mini → gpt-5-mini → (omit model param)

(omit model param) = call the task tool WITHOUT the model parameter. The platform uses its built-in default. This is the nuclear fallback — it always works.

Fallback rules:

• If the user specified a provider ("use Claude"), fall back within that provider only before hitting nuclear
• Never fall back UP in tier — a fast/cheap task should not land on a premium model
• Log fallbacks to the orchestration log for debugging, but never surface to the user unless asked

Passing the Model to Spawns

Pass the resolved model as the model parameter on every task tool call:

agent_type: "general-purpose"
model: "{resolved_model}"
mode: "background"
description: "{emoji} {Name}: {brief task summary}"
prompt: |
  ...

Only set model when it differs from the platform default (claude-sonnet-4.6). If the resolved model IS claude-sonnet-4.6, you MAY omit the model parameter — the platform uses it as default.

If you've exhausted the fallback chain and reached nuclear fallback, omit the model parameter entirely.

Spawn Output Format

When spawning, include the model in your acknowledgment:

🔧 Fenster (claude-sonnet-5) — refactoring auth module
🎨 Redfoot (claude-opus-4.8 · vision) — designing color system
📋 Scribe (claude-haiku-4.5 · fast) — logging session
⚡ Keaton (claude-opus-4.8 · bumped for architecture) — reviewing proposal
📝 McManus (claude-haiku-4.5 · fast) — updating docs

Include tier annotation only when the model was bumped or a specialist was chosen. Default-tier spawns just show the model name.

Valid Models

Premium: claude-opus-4.8, claude-opus-4.7, claude-opus-4.6 Standard: claude-sonnet-5, claude-sonnet-4.6, claude-sonnet-4.5, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-5.4, gpt-5.3-codex, gemini-2.5-pro Fast/Cheap: claude-haiku-4.5, gpt-5.4-mini, gpt-5-mini

Examples

Example 1: Backend dev writing API endpoints

• Role: Backend Dev
• Task: "implement REST endpoints for user management"
• Layer 3 decision: writing code → claude-sonnet-4.6 (standard tier)
• Spawn: 🔧 Fenster (claude-sonnet-4.6) — implementing user API endpoints

Example 2: User override

• User says: "use haiku for everything this session"
• Layer 1 overrides all other layers
• All spawns use claude-haiku-4.5 regardless of role or task

Example 3: Complex refactor

• Role: Backend Dev
• Task: "refactor 15 auth-related files to use new token system"
• Layer 3 base: claude-sonnet-4.5
• Task complexity: heavy multi-file refactor → switch to gpt-5.3-codex
• Spawn: 🔧 Fenster (gpt-5.3-codex · code specialist) — refactoring auth to new token system

Example 4: Scribe logging

• Role: Scribe
• Task: "log session to decisions.md"
• Layer 3: NOT writing code → claude-haiku-4.5
• Role mapping: Scribe always haiku, never bump
• Spawn: 📋 Scribe (claude-haiku-4.5 · fast) — logging session

Anti-Patterns

• ❌ Falling back UP in tier (fast task landing on premium model)
• ❌ Telling the user about fallback attempts ("Opus failed, trying Sonnet")
• ❌ Bumping Scribe or mechanical ops agents to higher tiers
• ❌ Using premium models for documentation or planning tasks
• ❌ Applying multiple complexity adjustments (cascading bumps)
• ❌ Forgetting to include model in spawn acknowledgment
• ❌ Downgrading vision-required tasks from opus