菜单
Select and create accurate software architecture, workflow, interaction, data, state, and deployment diagrams.
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
OpenAgentd workflow for version bumps, release PRs, GitHub releases, and release notes.
OpenAgentd workflow for keeping the compact feature catalogue and README aligned with user-visible changes.
Research first, propose a step-by-step implementation plan, and wait for explicit approval before writing any code.
Semantic search over the retained OpenAgentd feature catalogue, ADRs, and repository instructions using the turbovec experiment index.
OpenAgentd TDD workflow — write a failing test before the code that makes it pass, reproduce a bug with a test before fixing it. Use when implementing any logic, fixing any bug, or changing any existing behavior in backend (pytest) or frontend (Bun/RTL) code.
OpenAgentd testing reference — environment setup, run commands, and fix patterns for backend (pytest) and frontend (Bun/RTL). Load this for running, fixing, or adding coverage to existing tests. For writing a failing test before new code, use oad/test-driven-development instead.
| name | oad/diagram |
| description | Select and create accurate software architecture, workflow, interaction, data, state, and deployment diagrams. |
Use this skill to select a diagram type, then read only its reference.
| Request | Reference |
|---|---|
| System/runtime architecture, integrations, components | architecture |
| User, business, or operational process | workflow |
| Requests, events, streaming, or cross-service interactions | sequence |
| Lifecycle, retry, cancellation, or status transitions | state |
| Persistent records and relationships | data |
| Device, process, network, and trust topology | deployment |
Before drawing, state the one question it answers and verify facts in code,
tests, configuration, docs, and the nearest AGENTS.md. Use the smallest
useful diagram; split crowded views. Label relationships and directions, mark
proposals/unknowns, and keep open questions separate from facts. Do not expose
secrets or private data, or add a renderer/dependency unless explicitly asked.
