菜单
Answer natural-language questions about the architecture by cross-referencing ADRs, architecture docs, proposals, and the codebase. Designed for onboarding and architecture exploration.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Validate that a pull request follows principled conventions: references a plan or proposal, has correct labels, includes required sections in the description, and links to relevant documents. Use in CI or before merging to enforce PR quality.
Generate a structured pull request description from a DDD plan task. Reads the plan, task manifest, and implementation branch to produce a PR body with references to the originating proposal, plan, ADRs, and related tasks. Use when opening a PR for plan-driven work.
Process open GitHub issues through the principled pipeline. Lists untriaged issues, normalizes their metadata, ingests them into the documentation pipeline, and reports a summary. Operates as the batch entry point — delegates to /ingest-issue for each individual issue.
Implementation orchestration strategy for the Principled framework. Consult when working with DDD plan execution, git worktrees for task isolation, task manifests, or the decompose-spawn-validate-merge lifecycle. Covers worktree management, sub-agent patterns, task state machines, and manifest schema.
Top-level orchestrator for DDD plan execution. Decomposes a plan into tasks, iterates through phases respecting dependencies, spawns worktree-isolated agents, validates implementations, and merges results. Runs inline to coordinate multiple sub-agent spawns. Use for automated end-to-end plan execution.
Audit ADR coverage across modules. Identifies modules with no architectural governance, orphaned ADRs referencing removed modules, and stale architecture docs. Classifies findings by severity.
| name | arch-query |
| description | Answer natural-language questions about the architecture by cross-referencing ADRs, architecture docs, proposals, and the codebase. Designed for onboarding and architecture exploration. |
| allowed-tools | Read, Bash(git *), Bash(ls *), Bash(bash plugins/*) |
| user-invocable | true |
Answer natural-language questions about the architecture by searching across ADRs, architecture documents, proposals, and the codebase. Designed for onboarding and architecture exploration.
/arch-query "<question>"
| Argument | Required | Description |
|---|---|---|
"<question>" | Yes | A natural-language architecture question in quotes. |
Parse the question. Identify the query type:
Search relevant sources. Based on the query type:
docs/decisions/): Read titles, status, and body content. Focus on the "Decision" and "Context" sections.docs/architecture/): Read content for pattern descriptions, module relationships, and design explanations.docs/proposals/): Read for historical context on why features were built and what alternatives were considered.Synthesize the answer. Combine information from multiple sources into a coherent response:
Cite sources. Every claim should reference a specific document:
The plugin system uses a three-layer architecture (skills, hooks, foundation)
as described in docs/architecture/plugin-system.md. Module types are declared
in CLAUDE.md per ADR-003, and dependency direction rules are enforced
heuristically per ADR-014.
Suggest follow-ups. When relevant, suggest related queries or skills:
For a visual map of module governance, try: /arch-map
For a full coverage audit, try: /arch-audit
This skill is deliberately open-ended. It leverages Claude's ability to search, read, and synthesize rather than running deterministic scripts. The value is in connecting questions to the right documents, not in producing structured reports (that is what /arch-map and /arch-audit are for).