Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By 10XTeams
Orchestrate a structured PRD/TDD/test-case authoring, subsprint planning/execution, and ADO sync workflow. Includes a local Markdown wiki viewer/editor to browse and edit docs directly from the project folder.
npx claudepluginhub 10xteams/trio --plugin trioUse this subagent to turn a completed test report into a subsprint plan (`source: bugfix`). By default it auto-selects the latest report under `docs/Test-Report/` (the folder with the largest `<编号>-` prefix) — no user confirmation needed. The agent traces each failure back to the PRD (`docs/PRD/`), TDD (`docs/TDD/`), and actual code paths, classifies and prioritizes fixes, and writes the plan to `trio/subsprint/<编号>-MMDD-<word>/<编号>-subsprint-plan.md` with the required front-matter contract. Invoke via the `trio:subsprint-planner` skill. Optional input: a specific report folder name or path to override the auto-latest default.
Use this subagent to insert UI screenshots into a PRD document. It reads the target PRD, discovers the page/route list and test-account credentials from `docs/TDD/0.common/`, logs in via Playwright MCP, captures one screenshot per function/page referenced by the PRD, saves them under `docs/PRD/<module>/screenshots/`, and edits the PRD to embed each image next to the corresponding function. Invoke when the user wants to illustrate an existing PRD with real application screenshots. Required input: the PRD file path (absolute or repo-relative). Optional input: frontend URL override, explicit account identifier, or a specific function subset.
Use this subagent to write or update the detailed PRD for a single module. It reads `docs/PRD/PRD-Overview.md` and the relevant code, confirms the target module folder and sub-module list with the caller, writes the module overview (`0.<模块>-overview.md`), and writes per-sub-module files with a business-language Mermaid flow and a key-page function list. Sub-module candidates that are purely technical (session/token/middleware/guard/store/cache) are rejected. Required input: `module` — module name or folder (e.g. `1. 认证与登录模块`). Invoke via the `trio:prd-management` skill.
Use this subagent to cross-check the URS against all PRD documents and emit a structured gap report. It lists (1) features in PRD not in URS, (2) features in URS not in PRD, (3) PRD items marked "待实现 / TODO / Phase 2", and writes a JSON file to `trio/iteration/gap-check/Gap Check <n>-MMDD-<word>.json` — the `<n>` is the next integer following existing gap-check files. No arguments are required; always read the full URS and full PRD. Invoke via the `trio:prd-management` skill.
Use this subagent to write the PRD overview document (`docs/PRD/PRD-Overview.md`) in Simplified Chinese. It reads the URS and the existing code, proposes a module grouping, confirms the module list with the caller, creates numbered module folders under `docs/PRD/`, and writes the three-part overview (system overview, key concepts, per-module scope table). Invoke when a project is ready to start its PRD and has no overview yet, or when the existing overview needs to be rewritten from scratch. The caller is responsible for deciding when this is needed; this agent does not decide.
Use when starting a new Trio-managed project (needs the docs + trio skeleton plus the canonical `docs/TDD/0.common/tech-stack.md`, and `code-structure.md` if `code/` already exists). Produces the folder skeleton, root-level `CLAUDE.md` folder rule, the canonical `tech-stack.md`, and (when `code/` already has an implementation) dispatches `trio:tdd-code-structure` to produce `code-structure.md`.
Use when the user asks "what's next?", "下一步是什么", "what should I do now", or otherwise wants a recommendation for the next Trio action. Inspects the latest project state (test reports, subsprints, doc folders) and suggests a prioritized next step. This skill is advisory only — it does NOT dispatch agents or write files; the user invokes the recommended Skill themselves.
Orchestrate the PRD lifecycle for this project. Presents a decision tree (overview → by-module → URS gap check → screenshot augment), validates preconditions (URS exists, PRD-Overview done before by-module, etc.), and dispatches the correct agent with full context. Invoke whenever the user wants to write, update, or audit PRDs. The skill does not write PRDs itself — it orchestrates the `trio:prd-overview`, `trio:prd-by-module`, `trio:prd-check-urs-gap`, and `trio:prd-add-screenshot` agents.
Produce a subsprint plan file under `trio/subsprint/<n>-MMDD-<word>/<n>-subsprint-plan.md` from one of three sources — the latest failing test report (`trio:bugfix-plan`), a triaged URS-PRD gap report (`trio:urs-gap-plan`), or a brand-new fresh subsprint (`trio:subsprint-fresh-plan`). Validates preconditions, enforces the Plan Front-Matter Contract (`source`, `subsprint_id`, `subsprint_folder`, `docs_changes_expected`) and the Plan Checklist Contract (`Execute Coding` / `Update Docs` / `Update Test Case`). After the plan is generated, tells the user the recommended sequence: `/trio:subsprint-runner` (apply edits) → `/trio:subsprint-sync` (TC updates). Writes nothing itself. Invoke when the user wants to draft a plan — execution and sync are separate skills' jobs.
Execute a subsprint plan in place — pass the plan to `trio:subsprint-execute`, which applies each task's code-path and doc-path edits directly to `code/` and `docs/`. Resolves the plan input (absolute path, folder name, or bare `<n>`), parses front-matter, dispatches the executor, flips `Execute Coding` and `Update Docs` in the plan's checklist, and hands off to `/trio:subsprint-sync` for any remaining TC patches. Writes nothing itself. Invoke when the user wants to execute a plan produced by `/trio:subsprint-planner`.
Uses power tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Harness-native ECC plugin for engineering teams - 64 agents, 262 skills, 84 legacy command shims, reusable hooks, rules, MCP conventions, and operator workflows for Claude Code plus adjacent agent harnesses
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.
Comprehensive C4 architecture documentation workflow with bottom-up code analysis, component synthesis, container mapping, and context diagram generation
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
一个 Claude Code 插件,把 AI 编码从 "凭感觉写代码" 变成 "三大支柱 + 子冲刺闭环" 的可审计工作流。
Trio 是一套 AI coding 的 Harness 机制——但比起"让 AI 更会写代码",它更关注人在其中的过程:怎么让代码、文档、测试用例三者始终保持一致,而不是各跑各的。
工作流围绕三种文档("三大支柱")+ 一个执行单元展开:
配合契约文件和清单复选框,"AI 说做完了" 变成可机械验证的状态。
如果你和 AI 真正一起写过几周以上的代码,下面这些大概率都中过:
AI 没有 "为什么"。 没人写 PRD,AI 就只能 "看上去对" 地写代码——能跑、长得像,但和真实需求脱钩。一两个迭代之后开始累积漂移,没人说得清当初为什么这么做。
AI 找不到正确的文件。 跨模块改动时它会臆造路径、改错文件、把同一个东西在两处分别实现。src/... 不一定是 src/。
测试通过 ≠ 没 bug。 AI 写的测试很容易 "把现状当作期望"——失败的代码改一下,断言也跟着改一下,绿条永远在,bug 上线了才发现。
每次新对话从零开始。 上一次会话里达成的约定、走过的弯路、定下的命名,下一次会话全部消失。AI 重新猜,重新错。
大计划悄悄缩水。 AI 说 "我来做 X、Y、Z",最后只交付了 X,并宣布完成。没人能机械验证 "Z 到底有没有做"。
测试执行把上下文打爆。 一个 agent 跑 30+ 条用例,Playwright stdio 断连、context window 溢出、跑到一半就糊涂了,最后还是要人来兜底。
测试规格被悄悄削弱。 原本断言 "登录失败时显示账号被禁用",后来某次重写之后变成 "返回任何 4xx 即可"。回归测试还在绿,但实际保护已经没了。
Trio 针对的是这些流程问题,不是 "让 AI 更聪明"。
| 痛点 | Trio 的应对 |
|---|---|
| AI 没 "为什么" | 用 PRD 把"为什么"显式固化下来——先有整体再写模块,AI 做事之前必有依据 |
| 找不到正确文件 | 把项目的"代码地图"(前端路由 / 后端 endpoint / 模块到文件的对应)落成一份契约文档,所有后续动作都从它读起 |
| 测试 = 现状 | 测试用例必须显式列出验收点和禁止状态;后续修改默认只允许"加严",要放松必须走可见的全量重写 |
| 新对话从零开始 | 三大支柱 + 迭代历史本身就是冷启动地图——新会话直接读已有文档,不重新猜测 |
| 大计划悄悄缩水 | 每个子冲刺顶部都有一份固定结构的执行清单(代码 / 文档 / 测试用例),每个框只有唯一一个责任方能勾完;故意跳过必须写明原因。"做完了"是机械可读的状态,不是嘴上说说 |
| 测试执行打爆上下文 | 测试执行按子模块切片,每段结果立刻落盘成结构化文件;主控随时可以丢弃内存、重新读盘 |
| 规格被悄悄削弱 | 测试用例的修改默认只允许"加严"操作集;任何"删条目 / 放宽断言"在校验阶段就被拒绝,要放松必须走可见的全量重写 |
这个仓库是 个人版(Personal Edition)——开源(PolyForm Noncommercial 1.0.0),适合个人 / 研究 / 教学等非商业用途。
企业版(Enterprise Edition) 在个人版基础上面向团队协作场景额外提供三块能力,需要单独商业授权。
| 能力 | 个人版 | 企业版 |
|---|---|---|
| 三大支柱 + subsprint 闭环(PRD / TDD / Test-Case) | ✓ | ✓ |
/trio:next 智能路由 + Trio Stage 浏览器编辑器 | ✓ | ✓ |
| 测试用例 strengthen-only patch + 落盘 partials | ✓ | ✓ |
| Azure DevOps 任务与缺陷双向打通 | — | ✓ |
| Git Worktree 隔离 + Diff 校验 | — | ✓ |
| 合回团队分支时自动总结迭代内容 | — | ✓ |
把 ADO 上的 Work Item 直接拉进 Trio 的工作流:
/trio:subsprint-planner 多一种来源 ado——按 iteration / 标签从 DevOps 取任务和缺陷,自动生成 subsprint plan,task 与 Work Item 一一对应。Resolved / Closed,省掉手工同步。每次 subsprint 执行不在 code/ docs/ 主目录上动手,而是通过 Git Worktree 拉一份隔离副本:
code-diff.md / docs-diff.md,与 plan 里声明的 Code Paths / Doc Paths 做交叉校验——任何超出计划范围的改动会被高亮,等你确认。code/ docs/ 从未被污染,不存在"改错了不知道怎么恢复"的尴尬。当 subsprint 通过 worktree 校验、准备并入团队主分支时,基于 plan + diff + 测试报告自动生成迭代摘要:
可直接贴进 PR 描述、ADO 评论、站会更新——省掉人工写 release note 的时间。
联系仓库维护者(@10XTeams)获取商业授权与交付包。
Trio 把项目分成三个并行追踪的桶:
docs/ — 产品文档(PRD / TDD / Test-Case / Test-Report / URS)。独立 git 仓库(docs/.git)。code/ — 实现 + 配置。独立 git 仓库(code/.git)。trio/ — 工作流产物(子冲刺计划、URS-PRD gap-check JSON、subsprint notes)。故意不进 git——这些是过程产物,不属于产品文档历史。每个新项目第一步就是 /trio:init-project,它会创建这三个文件夹、写好根目录 CLAUDE.md 锁住三桶约定,并落下 tech-stack.md(必要时再生成 code-structure.md)。
| 文件 | 谁产出 | 谁读 |
|---|---|---|
docs/TDD/0.common/tech-stack.md | /trio:init-project | 几乎所有下游 skill |
docs/TDD/0.common/code-structure.md | trio:tdd-code-structure agent | PRD/TDD/TC/test 全部 |
trio/subsprint/<n>-MMDD-<word>/<n>-subsprint-plan.md | /trio:subsprint-planner | runner + sync |
docs/Test-Report/<n>-MMDD-<word>/bugs.json | /trio:test-management | bugfix planner + Trio Stage |
文件名是合同的一部分,不可重命名;schema(特别是 Bug schema v1.1)也是显式版本化的。
docs/ / code/、跑 Playwright、生成 PRD 内容的工作单元。这种分工的实际价值:当流程要调整时,只动 skill;当生成质量要提升时,只动 agent;不用每次改 prompt 都满地图找。
凡是会跑很久或可能被 auto-compact 截断的流程,Trio 都把中间结果写到磁盘:
partials/<sub#>.json主控 skill 在每一步重新从盘里读,从不依赖会话内存里 "我刚才看到的"。
每个 subsprint plan 顶部都是这三行(顺序、字面值都不能变):
## Execution Checklist
- [ ] **Execute Coding** — `/trio:subsprint-runner` (applies code edits in place)
- [ ] **Update Docs** — `/trio:subsprint-sync` (applies doc edits in place)
- [ ] **Update Test Case** — `/trio:tc-management` (patch or audit path)
规则:
— completed YYYY-MM-DD HH:MM by <skill>。[x] N/A 必须附一句原因,且不能再被翻回 [ ]。