菜单
Canonical architectural vocabulary — module, interface, seam, adapter, depth, leverage, locality. Use when reviewing architecture, deepening shallow modules, designing interfaces, or evaluating refactor candidates. NOT a domain glossary (use /ubiquitous-language for that).
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
Internal reference skill — coding standards and patterns for developer agents. Not user-invocable.
Close out work by committing local changes, pushing, waiting for green CI, and merging the PR. Universal — works on any branch with an open PR; no task structure required. Use when asked to 'merge the PR', 'ship it', 'merge and close', 'commit push merge', 'finalize the PR', 'land it on main', 'we're done — close it out', or whenever the user signals end-of-work and the next step is to merge. If (and only if) the current work is one phase of a multi-phase task in `tasks/**/phase-N-*/`, also make upcoming phase tech-decomposition documents accurate for the implementation that just landed by adding handoff notes or updating stale assumptions before merging — otherwise this step is silently skipped. Make sure to use this skill whenever the user signals end-of-implementation and wants the changes merged — even if they don't say the word "skill". NOT for opening a new PR (use plain `gh pr create`). NOT for addressing review comments (use /prc). NOT for preparing a session handoff to a fresh context (use /ph).
Use when a plan, design, or discovery document needs stress-testing for hidden assumptions, scope gaps, unresolved branches, or ambiguous wording; also when the user explicitly says "grill me".
Internal reference skill — shared conventions for all code review agents. Not user-invocable.
Implements a feature from a technical decomposition using TDD, keeping the task document updated and handing off cleanly to review. Use when starting, continuing, or resuming implementation from a tech-decomposition or task directory. Trigger: 'implement this task', 'start implementation', 'continue/resume implementation', 'build from the tech decomposition', or after /ct produces a decomposition. NOT for creating the plan (/ct), feature discovery (/nf), small untracked changes (/si-quick or /quick), or code review (/sr).
Use when asked to review code before merge or review a PR, branch, commit range, task path, or current working tree. Trigger on requests like 'review PR', 'review my changes', 'review this branch', 'is this ready to merge', or other pre-merge review requests. Supports task/spec-aware review when task docs exist. NOT for addressing review comments (use /prc).
| name | architecture-language |
| description | Canonical architectural vocabulary — module, interface, seam, adapter, depth, leverage, locality. Use when reviewing architecture, deepening shallow modules, designing interfaces, or evaluating refactor candidates. NOT a domain glossary (use /ubiquitous-language for that). |
| disable-model-invocation | true |
Shared vocabulary for every architectural conversation in this repo. Use these terms exactly — don't substitute "component", "service", "API", or "boundary". Consistent language is the whole point.
The full glossary lives in LANGUAGE.md. Load it when:
senior-architecture-reviewer, code-analysis)improve-codebase-architecture)A codebase full of shallow modules is hard for both humans and AI to navigate. The vocabulary in LANGUAGE.md makes the distinction between deep and shallow sayable, which is the prerequisite for fixing it.
When reviewing or planning, prefer the precise term over the casual one — say seam, not "boundary"; interface, not "API surface"; adapter, not "implementation of the interface".
The vocabulary and principles are adapted from Matt Pocock's improve-codebase-architecture/LANGUAGE.md (github.com/mattpocock/skills), which itself draws on Ousterhout (A Philosophy of Software Design) and Feathers (Working Effectively with Legacy Code).