* fix(3620): point the docs at files that actually exist docproof found 34 stale references; the reporter hand-read all 34 and reported the 8 that are real, explaining why the other 26 are deliberate (files the documents themselves label legacy or "superseded by", and one pre-Diataxis link label whose target still resolves). Those 26 are left alone — re-touching them would contradict the issue's own analysis. Every claim was re-verified against git ls-files at HEAD before editing. docs/INVENTORY.md said its roster is anchored by six drift-control tests. Five are gone (commands-doc-parity, agents-doc-parity, cli-modules-doc-parity, hooks-doc-parity in 5d8a8c4d; command-count-sync infbf30792), so the sentence now names the one that exists. Whether one test is sufficient coverage is a maintainer question the issue explicitly declined to answer, so no new drift tests are proposed here. The four translations were a revision further behind, each naming a seventh test deleted inae8bb707that the English file had already dropped. All four now match. Renamed targets corrected in CONTEXT.md, VERSIONING.md, docs/CONFIGURATION.md and the update workflow. The new test names carry no issue-NNN- prefix, which is what lint-regression-test-names requires, so they are the correct targets. docs/TESTING-SUITES.md is the one that could cost somebody time: it INSTRUCTED contributors to add an acknowledgment to the legacy drift-ack file, which CONTRIBUTING.md says to never use. Rewritten from the real workflow — per-PR fragments under the drift-acks directory, and a spent base-side ack is re-armed by rewording that fragment's reason in place, never by adding a duplicate, since two sources naming one path is a hard error. docs/skills/discovery-contract.md's heading named a query module deleted in11918dcc. The section was REMOVED rather than retargeted: its documented behavior (skip the deprecated root) is not what the surviving code does — skill-manifest includes that root marked deprecated:true — so retargeting would have documented something false. Found and fixed inline, same class: VERSIONING.md described an SDK bundling step the release workflow does not have (zero such mentions in that file); CONFIGURATION.md and four translations named a dead model-catalog triple collapsed by ADR-457. Dead config removed: the changeset lint's user-facing prefix list still carried two retired sdk entries. git ls-files -- 'sdk/*' returns nothing. No test pins that array. Left deliberately: the comment explaining the retired catalog path, the install regression test that reconstructs the old broken layout to prove it fails, and the generated test-timings cache. Each is a legitimate mention of a dead path, not drift. Note lint-removed-but-needed cannot catch this class: it diffs baseRef...HEAD, so it only sees files deleted in the change under review. These were orphaned by PRs that predate the lint. A repo-wide existence audit would need a suppression mechanism for the 26 deliberate mentions above; that is a feature, not part of this fix. Fixes #3620 * chore(3620): backfill changeset PR number (#3658) --------- Co-authored-by: sim <sim@local>
220 lines
7.8 KiB
Markdown
220 lines
7.8 KiB
Markdown
# 如何配置模型配置文件
|
||
|
||
为您的项目选择合适的模型层级策略,然后在不编写大型覆盖块的情况下调整单个代理或整个阶段类型。本指南从最简单的控制选项开始,逐步介绍到动态路由。
|
||
|
||
---
|
||
|
||
## 四种配置文件(以及 `adaptive` 和 `inherit`)
|
||
|
||
在 `.planning/config.json` 中设置 `model_profile`,或通过 `/gsd-config --profile <name>` 设置:
|
||
|
||
| 配置文件 | 规划器 | 执行器 | 研究员 | 验证器 | 适用场景 |
|
||
|---------|---------|----------|-------------|----------|----------|
|
||
| `quality` | Opus | Opus | Opus | Sonnet | 对成本要求较低、注重生产质量的工作 |
|
||
| `balanced` | Opus | Sonnet | Sonnet | Sonnet | 常规开发——默认选项 |
|
||
| `budget` | Sonnet | Sonnet | Haiku | Haiku | 快速原型开发、成本敏感场景 |
|
||
| `adaptive` | Opus | Sonnet | Sonnet | Sonnet | 与其他层级在运行时感知配置文件下的解析方式相同;在频繁切换运行时环境时使用 |
|
||
| `inherit` | (会话模型) | (会话模型) | (会话模型) | (会话模型) | 非 Anthropic 提供商(OpenRouter、本地模型)——所有代理遵循当前会话模型 |
|
||
|
||
上表展示的是代表性子集。全部 33 个内置代理在 `gsd-core/bin/shared/model-catalog.json` 中均有明确的按配置文件层级分配。完整表格请参阅配置参考中的 [模型配置文件](../CONFIGURATION.md#model-profiles)。
|
||
|
||
**通过命令快速切换:**
|
||
|
||
```bash
|
||
/gsd-config --profile balanced # Normal development
|
||
/gsd-config --profile budget # Prototyping or high-cost phases
|
||
/gsd-config --profile quality # Production release
|
||
/gsd-config --profile inherit # OpenRouter, local models
|
||
```
|
||
|
||
**或直接编辑 `.planning/config.json`:**
|
||
|
||
```json
|
||
{
|
||
"model_profile": "balanced"
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
## 按代理覆盖(`model_overrides`)
|
||
|
||
如果某个代理需要不同的层级而不想更改整个配置文件,请使用 `model_overrides`:
|
||
|
||
```json
|
||
{
|
||
"model_profile": "balanced",
|
||
"model_overrides": {
|
||
"gsd-executor": "opus",
|
||
"gsd-codebase-mapper": "haiku"
|
||
}
|
||
}
|
||
```
|
||
|
||
有效值:`opus`、`sonnet`、`haiku`、`inherit`,或任何完全限定的模型 ID(例如 `"openai/o3"`、`"google/gemini-2.5-pro"`)。
|
||
|
||
`model_overrides` 可在 `.planning/config.json` 中按项目设置,也可在 `~/.gsd/defaults.json` 中全局设置。项目级条目在冲突时优先;不冲突的全局条目会被保留。
|
||
|
||
**关于 Codex 和 OpenCode 的重要说明:** 这些运行时会在安装时将解析后的模型嵌入每个代理的静态配置中。编辑 `model_overrides` 后,需重新运行安装程序使更改生效:
|
||
|
||
```bash
|
||
npx @opengsd/gsd-core@latest --codex --global # or --opencode, --kilo, etc.
|
||
```
|
||
|
||
---
|
||
|
||
## 按阶段类型设置模型(`models`)
|
||
|
||
如果您希望在不学习全部 33 个代理名称的情况下实现"规划阶段用 Opus、其余用 Sonnet"的效果,请使用 `models` 块。它将六种阶段类型映射到层级别名:
|
||
|
||
```json
|
||
{
|
||
"model_profile": "balanced",
|
||
"models": {
|
||
"planning": "opus",
|
||
"discuss": "opus",
|
||
"research": "sonnet",
|
||
"execution": "opus",
|
||
"verification": "sonnet",
|
||
"completion": "sonnet"
|
||
}
|
||
}
|
||
```
|
||
|
||
阶段类型及其对应的代理:
|
||
|
||
| 阶段类型 | 涵盖的代理 |
|
||
|---|---|
|
||
| `planning` | `gsd-planner`、`gsd-roadmapper`、`gsd-pattern-mapper` |
|
||
| `research` | `gsd-phase-researcher`、`gsd-project-researcher`、`gsd-research-synthesizer`、`gsd-codebase-mapper`、`gsd-ui-researcher` |
|
||
| `execution` | `gsd-executor`、`gsd-debugger`、`gsd-doc-writer` |
|
||
| `verification` | `gsd-verifier`、`gsd-plan-checker`、`gsd-integration-checker`、`gsd-nyquist-auditor`、`gsd-ui-checker`、`gsd-ui-auditor`、`gsd-doc-verifier`、`gsd-code-reviewer` |
|
||
| `discuss` | `gsd-assumptions-analyzer` |
|
||
| `completion` | 保留——目前无子代理;已被模式接受以备向后兼容 |
|
||
|
||
`models` 块仅接受层级别名(`opus`、`sonnet`、`haiku`、`inherit`)。如需使用完全限定的模型 ID,请改用按代理设置的 `model_overrides`。
|
||
|
||
**将 `models` 与按代理例外结合使用:**
|
||
|
||
```json
|
||
{
|
||
"model_profile": "balanced",
|
||
"models": {
|
||
"research": "sonnet"
|
||
},
|
||
"model_overrides": {
|
||
"gsd-codebase-mapper": "haiku"
|
||
}
|
||
}
|
||
```
|
||
|
||
全部五个研究代理解析为 `sonnet`,*除* `gsd-codebase-mapper` 被固定为 `haiku` 之外。
|
||
|
||
---
|
||
|
||
## 动态路由——默认使用低成本层级,失败时升级
|
||
|
||
如果您希望默认使用较低成本的层级,仅在代理未通过质量门控时才升级,请启用 `dynamic_routing`:
|
||
|
||
```json
|
||
{
|
||
"dynamic_routing": {
|
||
"enabled": true,
|
||
"tier_models": {
|
||
"light": "haiku",
|
||
"standard": "sonnet",
|
||
"heavy": "opus"
|
||
},
|
||
"escalate_on_failure": true,
|
||
"max_escalations": 1
|
||
}
|
||
}
|
||
```
|
||
|
||
每个代理都有一个默认层级(`light`、`standard` 或 `heavy`)。第一次尝试时,GSD Core 选择 `tier_models[default_tier]`。如果编排器检测到软失败(验证不确定、计划检查被标记等),则将代理提升一级重新启动。`max_escalations` 限制总重试次数。
|
||
|
||
已处于 `heavy` 层级的代理无法进一步升级。
|
||
|
||
**在保留动态解析的同时关闭升级:**
|
||
|
||
```json
|
||
{
|
||
"dynamic_routing": {
|
||
"enabled": true,
|
||
"escalate_on_failure": false
|
||
}
|
||
}
|
||
```
|
||
|
||
无论结果如何,每次尝试都使用 `tier_models[default_tier]`——适用于希望明确指定层级到模型的映射但不需要升级行为的场景。
|
||
|
||
`dynamic_routing` **默认禁用**。省略该块或设置 `enabled: false` 将保留静态解析。
|
||
|
||
---
|
||
|
||
## 在非 Anthropic 运行时上使用 GSD Core
|
||
|
||
如果您为 Codex、OpenCode、Gemini CLI 或 Kilo 安装了 GSD Core,安装程序已在您的配置中设置了 `resolve_model_ids: "omit"`。这告知 GSD Core 跳过 Anthropic 模型 ID 解析,让运行时选择其自己的默认模型。基本情况下无需手动设置。
|
||
|
||
**如果您希望在 Codex 上使用分层模型:**
|
||
|
||
```json
|
||
{
|
||
"runtime": "codex",
|
||
"model_profile": "balanced"
|
||
}
|
||
```
|
||
|
||
GSD Core 将每个层级别名解析为运行时层级映射中定义的 Codex 原生模型和推理力度。
|
||
|
||
**如果您希望在任意非 Claude 运行时上使用按代理模型 ID:**
|
||
|
||
```json
|
||
{
|
||
"resolve_model_ids": "omit",
|
||
"model_overrides": {
|
||
"gsd-planner": "o3",
|
||
"gsd-executor": "o4-mini",
|
||
"gsd-debugger": "o3"
|
||
}
|
||
}
|
||
```
|
||
|
||
有关完整的运行时感知配置文件参考及 `model_policy` 接口(v1.42 中新增的提供商中立预设),请参阅[配置参考——模型配置文件](../CONFIGURATION.md#model-profiles)。
|
||
|
||
---
|
||
|
||
## 解析优先级(从高到低)
|
||
|
||
当多个层级同时适用时,解析器选取优先级最高的条目:
|
||
|
||
```text
|
||
1. model_overrides[<agent>] — per-agent; full IDs; targeted exception
|
||
2. dynamic_routing.tier_models[<tier>] — when enabled; escalates on soft failure
|
||
3. models[<phase_type>] — coarse phase-level tier
|
||
4. model_profile (per-agent column) — global tier strategy
|
||
5. Runtime default — when nothing else applies
|
||
```
|
||
|
||
---
|
||
|
||
## 选择合适的控制选项
|
||
|
||
| 您的需求 | 使用 |
|
||
|---|---|
|
||
| 对所有代理采用统一的层级策略 | `model_profile` |
|
||
| 粗粒度的阶段级调整("规划阶段用 Opus") | `models.<phase_type>` |
|
||
| 按代理精细控制("强制代码库映射器使用 Haiku") | `model_overrides[<agent>]` |
|
||
| 为特定代理指定完全限定的模型 ID | `model_overrides[<agent>]: "openai/gpt-5"` |
|
||
| 默认低成本,仅在失败时升级 | `dynamic_routing` |
|
||
| 所有代理遵循会话模型(非 Anthropic 提供商) | `model_profile: "inherit"` |
|
||
|
||
---
|
||
|
||
## 相关文档
|
||
|
||
- [配置参考](../CONFIGURATION.md)
|
||
- [多代理编排](../explanation/multi-agent-orchestration.md)
|
||
- [命令参考](../COMMANDS.md)
|
||
- [文档索引](../README.md)
|