| 1 | # Model Selection |
| 2 | |
| 3 | > Determines which LLM model to use for each agent spawn. |
| 4 | |
| 5 | ## SCOPE |
| 6 | |
| 7 | ✅ THIS SKILL PRODUCES: |
| 8 | - A resolved `model` parameter for every `task` tool call |
| 9 | - Persistent model preferences in `.squad/config.json` |
| 10 | - Spawn acknowledgments that include the resolved model |
| 11 | |
| 12 | ❌ THIS SKILL DOES NOT PRODUCE: |
| 13 | - Code, tests, or documentation |
| 14 | - Model performance benchmarks |
| 15 | - Cost reports or billing artifacts |
| 16 | |
| 17 | ## Context |
| 18 | |
| 19 | Squad supports 18+ models across three tiers (premium, standard, fast). The coordinator must select the right model for each agent spawn. Users can set persistent preferences that survive across sessions. |
| 20 | |
| 21 | ## 5-Layer Model Resolution Hierarchy |
| 22 | |
| 23 | Resolution is **first-match-wins** — the highest layer with a value wins. |
| 24 | |
| 25 | | Layer | Name | Source | Persistence | |
| 26 | |-------|------|--------|-------------| |
| 27 | | **0a** | Per-Agent Config | `.squad/config.json` → `agentModelOverrides.{name}` | Persistent (survives sessions) | |
| 28 | | **0b** | Global Config | `.squad/config.json` → `defaultModel` | Persistent (survives sessions) | |
| 29 | | **1** | Session Directive | User said "use X" in current session | Session-only | |
| 30 | | **2** | Charter Preference | Agent's `charter.md` → `## Model` section | Persistent (in charter) | |
| 31 | | **3** | Task-Aware Auto | Code → sonnet, docs → haiku, visual → opus | Computed per-spawn | |
| 32 | | **4** | Default | `claude-haiku-4.5` | Hardcoded fallback | |
| 33 | |
| 34 | **Key principle:** Layer 0 (persistent config) beats everything. If the user said "always use opus" and it was saved to config.json, every agent gets opus regardless of role or task type. This is intentional — the user explicitly chose quality over cost. |
| 35 | |
| 36 | ## AGENT WORKFLOW |
| 37 | |
| 38 | ### On Session Start |
| 39 | |
| 40 | 1. READ `.squad/config.json` |
| 41 | 2. CHECK for `defaultModel` field — if present, this is the Layer 0 override for all spawns |
| 42 | 3. CHECK for `agentModelOverrides` field — if present, these are per-agent Layer 0a overrides |
| 43 | 4. STORE both values in session context for the duration |
| 44 | |
| 45 | ### On Every Agent Spawn |
| 46 | |
| 47 | 1. CHECK Layer 0a: Is there an `agentModelOverrides.{agentName}` in config.json? → Use it. |
| 48 | 2. CHECK Layer 0b: Is there a `defaultModel` in config.json? → Use it. |
| 49 | 3. CHECK Layer 1: Did the user give a session directive? → Use it. |
| 50 | 4. CHECK Layer 2: Does the agent's charter have a `## Model` section? → Use it. |
| 51 | 5. CHECK Layer 3: Determine task type: |
| 52 | - Code (implementation, tests, refactoring, bug fixes) → `claude-sonnet-4.6` |
| 53 | - Prompts, agent designs → `claude-sonnet-4.6` |
| 54 | - Visual/design with image analysis → `claude-opus-4.6` |
| 55 | - Non-code (docs, planning, triage, changelogs) → `claude-haiku-4.5` |
| 56 | 6. FALLBACK Layer 4: `claude-haiku-4.5` |
| 57 | 7. INCLUDE model in spawn acknowledgment: `🔧 {Name} ({resolved_model}) — {task}` |
| 58 | |
| 59 | ### When User Sets a Preference |
| 60 | |
| 61 | **Trigger phrases:** "always use X", "use X for everything", "switch to X", "default to X" |
| 62 | |
| 63 | 1. VALIDATE the model ID against the catalog (18+ models) |
| 64 | 2. WRITE `defaultModel` to `.squad/config.json` (merge, don't overwrite) |
| 65 | 3. ACKNOWLEDGE: `✅ Model preference saved: {model} — all future sessions will use this until changed.` |
| 66 | |
| 67 | **Per-agent trigger:** "use X for {agent}" |
| 68 | |
| 69 | 1. VALIDATE model ID |
| 70 | 2. WRITE to `agentModelOverrides.{agent}` in `.squad/config.json` |
| 71 | 3. ACKNOWLEDGE: `✅ {Agent} will always use {model} — saved to config.` |
| 72 | |
| 73 | ### When User Clears a Preference |
| 74 | |
| 75 | **Trigger phrases:** "switch back to automatic", "clear model preference", "use default models" |
| 76 | |
| 77 | 1. REMOVE `defaultModel` from `.squad/config.json` |
| 78 | 2. ACKNOWLEDGE: `✅ Model preference cleared — returning to automatic selection.` |
| 79 | |
| 80 | ### STOP |
| 81 | |
| 82 | After resolving the model and including it in the spawn template, this skill is done. Do NOT: |
| 83 | - Generate model comparison reports |
| 84 | - Run benchmarks or speed tests |
| 85 | - Create new config files (only modify existing `.squad/config.json`) |
| 86 | - Change the model after spawn (fallback chains handle runtime failures) |
| 87 | |
| 88 | ## Config Schema |
| 89 | |
| 90 | `.squad/config.json` model-related fields: |
| 91 | |
| 92 | ```json |
| 93 | { |
| 94 | "version": 1, |
| 95 | "defaultModel": "claude-opus-4.6", |
| 96 | "agentModelOverrides": { |
| 97 | "fenster": "claude-sonnet-4.6", |
| 98 | "mcmanus": "claude-haiku-4.5" |
| 99 | } |
| 100 | } |
| 101 | ``` |
| 102 | |
| 103 | - `defaultModel` — applies to ALL agents unless overridden by `agentModelOverrides` |
| 104 | - `agentModelOverrides` — per-agent overrides that take priority over `defaultModel` |
| 105 | - Both fields are optional. When absent, Layers 1-4 apply normally. |
| 106 | |
| 107 | ## Fallback Chains |
| 108 | |
| 109 | If a model is unavailable (rate limit, plan restriction), retry within the same tier: |
| 110 | |
| 111 | ``` |
| 112 | Premium: claude-opus-4.6 → claude-opus-4.6-fast → claude-opus-4.5 → claude-sonnet-4.6 |
| 113 | Standard: claude-sonnet-4.6 → gpt-5.4 → claude-sonnet-4.5 → gpt-5.3-codex → claude-sonnet-4 |
| 114 | Fast: claude-haiku-4.5 → gpt-5.1-codex-mini → gpt-4.1 → gpt-5-mini |
| 115 | ``` |
| 116 | |
| 117 | **Never fall UP in tier.** A fast task won't land on a premium model via fallback. |