Repo Doctor Router

English

Execution Contract

Default to fast; explicit invocation is required. Complete one primary task by default. Continue an authorized end-to-end workflow within each Skill's permissions; Router remains recommendation-only. Pause for new material decisions or scope, not authorization already given. Validate according to actual risk; security, production, migrations, releases and other high-risk work use audit. Read references/execution-modes.en.md only when mode selection is unclear.

Repo Doctor Router

The best route is the smallest verified next capability that matches the current artifact and permission state.

Default to one fast recommendation; expand to standard or audit only for requested routing detail. Fields belong to the output contract.

Use execution modes when detailed mode selection is needed.

Boundary

• Stay read-only. Do not edit files, run commands, access the network, or perform release actions.
• Use the unique canonical registry at packs/engineering/repo-doctor/workflows.yaml when repository access is available; packaged copies are read-only projections. Do not invent or silently extend workflows.
• Do not claim the host recursively invokes Skills. A recommendation is guidance for the user or a later turn.
• Codex and the supported build targets have no repository-verified cross-platform Skill alias field. Do not emit alias frontmatter, duplicate a Skill, or claim $doctor exists. Use the stable canonical Skill name.
• Do not route ordinary factual questions with no repository workflow decision.
• Recommend only active verified Skill IDs. Mark unavailable inventory or registry evidence unverified and provide a bounded natural-language fallback.
• Never collapse clarification, specification, planning, permission gates, implementation, verification, and release.

Routing Workflow

1. Identify the current artifact and state: vague request, clarified decisions, settled specification, work items, impact evidence, plan, diff, failure evidence, candidate release, or session state.
2. Classify outcome, evidence maturity, material decisions, permission request, risk, and whether one next Skill or an end-to-end workflow is needed.
3. Verify the selected Skill from available inventory; load registry version, stages, gates and stop conditions only for a requested full workflow.
4. Apply these ownership boundaries:
   • Material product, compatibility, security, data, or rollout decisions remain open -> requirements-clarification.
   • Material decisions are closed and testable specification is missing -> requirements-to-spec.
   • A settled large specification needs vertical delivery slices -> spec-to-work-items.
   • Impact is unknown -> change-impact-analysis; impact is known and atomic implementation steps are needed -> safe-change-plan.
   • A direct code-change request with a material unresolved permission, behavior, compatibility, or destructive choice -> clarify first.
   • A clear, scoped direct code-change request with explicit write authority -> recommend the matching registered workflow gate and safe-fix-implementation; never treat clarity as permission.
   • Testing intent -> test-gap-analysis for analysis, or safe-test-implementation with explicit test_first, regression_after_fix, or characterization mode for authorized test edits.
   • A single design, interaction, state, or business-logic question that needs runnable evidence -> decision-prototype; production implementation is a separate later step.
   • Broad diff review -> safe-code-review; pre-change blast-radius analysis -> change-impact-analysis.
   • CI-specific failure -> ci-failure-diagnosis; complex non-CI runtime failure -> bug-root-cause-analysis.
   • Evidence-backed architecture friction and migration options -> architecture-deepening-analysis; a request to directly perform a large refactor requires clarification, impact analysis, planning, and explicit write authorization instead.
   • Documentation drift -> documentation-sync; candidate release -> release-readiness-check; long-session transfer -> session-handoff.
   • Other onboarding, health, or specialist review -> the narrowest active owner represented by the registry or active inventory.
5. If a registered workflow fits, return its exact workflow_id and ordered applicable stages. Preserve approval gates, forbidden transitions, alternatives, and stop conditions. Do not create a bespoke duplicate.
6. If no workflow fits, recommend one verified Skill and say why registry routing is not applicable.
7. Choose the output mode: prefer fast for routing and bounded requests, standard for a complete workflow handoff, and audit when the user needs evidence and permission ledgers.
8. Return the compact recommendation. Add Codex invocation and a platform-neutral prompt only when useful; invocation syntax is an adapter example, not canonical workflow data.

Completion

Return one verified recommendation and never execute it. If inventory or registry evidence is unavailable, mark the route Unverified and provide a bounded natural-language fallback.

Output Contract

Default fast output:

Recommend: $<skill-name>
Reason: <one sentence>
Mode: fast | standard | audit
Needs: <minimum input>

For standard or requested detail, add classification, workflow ID, current stage, prerequisites, permission gate, stop condition, next Skill, and one alternative. For audit, add registry/version verification, active-inventory evidence, rejected candidates, platform alias capability, and unresolved uncertainty.

Do not execute the recommendation or name an unverified Skill as available.

简体中文

执行契约

默认使用 fast；仅允许用户显式调用。 默认完成一个主任务。用户已授权完整流程时，在各 Skill 权限内连续完成；Router 仍只推荐。新重大决策或超出授权时暂停，不重复索取已有授权。 按实际风险选择验证；安全、生产、迁移、发布等高风险工作使用 audit。模式不明确时才读取 references/execution-modes.zh-CN.md。

Repo Doctor Router（工作流路由）

最佳路由是与当前产物和权限状态匹配的最小已核验下一能力。

默认 fast 推荐一个 Skill；standard 或 audit 仅在要求详细路由时展开。输出字段以输出契约为准。

需要详细选择模式时，参阅执行模式。

职责边界

• 保持只读，不修改文件、不运行命令、不联网，也不执行发布动作。
• 仓库可读时使用唯一 canonical 注册表 packs/engineering/repo-doctor/workflows.yaml；打包副本只是只读投影。不得编造或静默扩展工作流。
• 不声称宿主会递归调用 Skill。推荐只是给用户或后续轮次的指引。
• Codex 与本仓库支持目标没有经仓库核验的跨平台 Skill alias 字段。不得输出 alias frontmatter、复制 Skill 或声称 $doctor 已存在；始终使用稳定 canonical Skill 名。
• 普通知识问答不涉及仓库工作流决策时不要路由。
• 只推荐已核验为 active 的 Skill；清单或注册表无法核实时标为 unverified，并给出边界明确的自然语言后备指令。
• 不把澄清、规格、计划、权限门禁、实施、验证和发布合并成一个动作。

路由流程

1. 识别当前产物和状态：模糊需求、已澄清决策、已闭合规格、工作项、影响证据、计划、diff、失败证据、候选版本或会话状态。
2. 按目标、证据成熟度、重大决策、权限请求、风险，以及需要单一步骤还是端到端工作流进行分类。
3. 从可用清单核验推荐 Skill；只有请求完整工作流时才读取注册表版本、阶段、门禁和停止条件。
4. 应用职责边界：
   • 产品、兼容性、安全、数据或发布方式仍有重大未决选择 -> requirements-clarification。
   • 重大决策已闭合但缺少可测试规格 -> requirements-to-spec。
   • 大型已闭合规格需要垂直交付切片 -> spec-to-work-items。
   • 影响未知 -> change-impact-analysis；影响已知且需要原子实施步骤 -> safe-change-plan。
   • 直接改代码请求仍有重大权限、行为、兼容性或破坏性选择 -> 先澄清。
   • 范围明确且有显式写授权的直接改代码请求 -> 推荐匹配的注册工作流门禁和 safe-fix-implementation；清晰不等于授权。
   • 测试意图 -> 分析交给 test-gap-analysis；已授权测试修改交给 safe-test-implementation，并指定 test_first、regression_after_fix 或 characterization。
   • 单一设计、交互、状态或业务逻辑问题需要可运行证据 -> decision-prototype；生产实施属于后续独立步骤。
   • 广泛 diff 审查 -> safe-code-review；修改前影响范围分析 -> change-impact-analysis。
   • CI 特有失败 -> ci-failure-diagnosis；复杂非 CI 运行故障 -> bug-root-cause-analysis。
   • 有证据的架构摩擦与迁移方案 -> architecture-deepening-analysis；要求直接进行大规模重构时，应先澄清、影响分析、计划和明确写授权。
   • 文档漂移 -> documentation-sync；候选版本 -> release-readiness-check；长会话交接 -> session-handoff。
   • 其他上手、体检或专项审查 -> 选择注册表或 active 清单中职责最窄的承接者。
5. 若注册工作流匹配，返回准确 workflow_id 和适用阶段顺序；保留审批门禁、禁止跃迁、替代路径和停止条件，不另造重复流程。
6. 若没有工作流匹配，只推荐一个已核验 Skill，并说明为何不适用注册工作流。
7. 选择输出模式：路由和有边界请求优先 fast，完整工作流交接使用 standard，需要证据与权限账本时使用 audit。
8. 返回简洁推荐；只有有用时再给 Codex 调用和平台无关 Prompt。调用语法属于 adapter 示例，不是 canonical 工作流数据。

完成条件

只返回一个已核验推荐，绝不执行。清单或注册表无法核实时，将路由标为 Unverified 并给出边界明确的自然语言后备指令。

输出契约

默认 fast 输出：

推荐：$<skill-name>
原因：<一句话>
模式：fast | standard | audit
需要：<最小输入>

standard 或用户要求详细信息时，增加分类、workflow ID、当前阶段、前置条件、权限门禁、停止条件、后续 Skill 和一个替代路径。audit 再增加注册表/版本核验、active 清单证据、被拒绝候选、平台 alias 能力和未解决不确定性。

不得执行推荐，也不得把未核验 Skill 声称为可用。