--- name: configure-model description: "交互式配置 AI 模型提供商、模型 ID 和每个 sub-agent 的角色分配。" type: procedure applicable_to: - all inputs: [] outputs: [] linked_skills: [] linked_rules: [] linked_workflows: [] owner: Kucell last_verified: 2026-08-06 status: stable --- # 🤖 Model Configuration Workflow (/configure-model) Guides you through setting up your AI provider, model aliases, and role assignments, then automatically updates `config/reasoning-config.yml` and all sub-agent files. --- ## Step 1: Read Current Configuration First, I'll read the existing config and display the current state: 1. Read `.agent/config/reasoning-config.yml` 2. Read the `model:` field from all sub-agents: - `sub-agents/planner.md` - `sub-agents/implementer.md` - `sub-agents/researcher.md` - `sub-agents/code-reviewer.md` - `sub-agents/documenter.md` - `sub-agents/entropy-scanner.md` - `sub-agents/session-manager.md` 3. Display a current config summary: ``` Current Configuration ────────────────────────────── Provider: {api.provider} API Key Env: {api.api_key_env} Model Aliases: fast → {models.fast} standard → {models.standard} premium → {models.premium} Role Assignments: planner → {roles.planner} ({resolved model ID}) implementer → {roles.implementer} ({resolved model ID}) researcher → {roles.researcher} ({resolved model ID}) code_reviewer → {roles.code_reviewer} ({resolved model ID}) documenter → {roles.documenter} ({resolved model ID}) entropy_scanner → {roles.entropy_scanner} ({resolved model ID}) Cost Mode: {active_mode or "per-role"} ────────────────────────────── ``` --- ## Step 2: Choose AI Provider **Select your AI provider:** - [ ] **Anthropic** (default) — Claude model family - [ ] **OpenAI** — GPT / o1 models - [ ] **Azure OpenAI** — Azure-hosted OpenAI models - [ ] **Ollama** — Local models (qwen / llama / mistral, etc.) - [ ] **Custom** — Any OpenAI-compatible API provider - [ ] Keep current setting For non-Anthropic providers, also provide: - `base_url`: API endpoint (`http://localhost:11434` for Ollama; required for Azure / Custom) - `api_key_env`: Environment variable name holding your API key (e.g. `OPENAI_API_KEY`) --- ## Step 3: Configure Model Aliases **Specify the real model ID for each alias:** | Alias | Used For | Current | New | |-------|----------|---------|-----| | `fast` | Lightweight, high-frequency tasks (documenter, entropy-scanner) | {current} | | | `standard` | Everyday development (implementer, researcher, reviewer) | {current} | | | `premium` | Complex reasoning & architecture (planner) | {current} | | **Common model IDs:** ``` Anthropic: claude-haiku-4-5-20251001 / claude-sonnet-4-6 / claude-opus-4-6 OpenAI: gpt-4o-mini / gpt-4o / o1 Azure: Same as OpenAI; use your deployment name Ollama: qwen2.5-coder:7b / qwen2.5-coder:32b / llama3.1:70b ``` > Press Enter to skip and keep the current value. --- ## Step 4: Adjust Role Assignments (Optional) **Do any roles need a different alias or model ID?** Default assignments: | Role | Default Alias | Notes | |------|--------------|-------| | `planner` | `premium` | Planning needs strongest reasoning | | `implementer` | `standard` | Code execution | | `researcher` | `standard` | Research and feasibility | | `code_reviewer` | `standard` | Code review | | `documenter` | `fast` | Docs generation | | `entropy_scanner` | `fast` | High-frequency, cost-sensitive | To override a role, specify the role name and a new alias (`fast` / `standard` / `premium`) or a full model ID. --- ## Step 5: Cost Mode Override (Optional) **Enable a global cost mode to override all role assignments?** - [ ] **Disabled** (leave empty — use per-role assignments from Step 4) - [ ] **conservative** — All roles use `fast`, lowest cost - [ ] **balanced** — All roles use `standard`, everyday default - [ ] **quality** — planner + verification use `premium`, rest use `standard` > When enabled, the cost mode overrides Step 4 role assignments. --- ## 🤖 My Actions After collecting your answers, I will: ### 1. Update `config/reasoning-config.yml` - Write new `api.provider` / `api.base_url` / `api.api_key_env` - Update `models.fast` / `models.standard` / `models.premium` - Update `roles` assignments - Update `active_mode` ### 2. Sync `model:` in all sub-agent files Resolve each alias to its actual model ID and update: | Sub-agent File | Updated To | |---------------|-----------| | `sub-agents/planner.md` | `model: {resolved model ID for roles.planner}` | | `sub-agents/implementer.md` | `model: {resolved model ID for roles.implementer}` | | `sub-agents/researcher.md` | `model: {resolved model ID for roles.researcher}` | | `sub-agents/code-reviewer.md` | `model: {resolved model ID for roles.code_reviewer}` | | `sub-agents/documenter.md` | `model: {resolved model ID for roles.documenter}` | | `sub-agents/entropy-scanner.md` | `model: {resolved model ID for roles.entropy_scanner}` | ### 3. Output Change Summary ``` ✅ Model configuration updated ────────────────────────────── Provider: {new value} API Key Env: {new value} Model Alias Changes: fast {old} → {new} standard {old} → {new} premium {old} → {new} Sub-agent Model Changes: planner {old} → {new} implementer {old} → {new} ... (only changed roles shown) Cost Mode: {new value} ────────────────────────────── ⚠️ Reminder: make sure {api.api_key_env} is set in your shell environment. ``` --- ## ⚖️ 校准提示（Recalibration Tip） > 切换 provider/model 后，建议运行 `/recalibrate` 重测基线。 - **suggestion only**：仅为提示，不自动运行、不自动触发任何模型切换或配置升级。 - 目的：换模型后能力边界可能漂移，用同一固化代表工作流（`/ship` REVIEW→COMMIT 链路或 最小可复现 benchmark 子集）重跑，产出质量/成本/时长三维对比，写入 `metrics/model-calibration.json`（模型维度唯一事实源）。 - 相关：`.agent/workflows/recalibrate.md`（持续校准闭环，P-003）。