菜单
Use this skill to start a new MCS agent project. Creates the folder structure in Build-Guides/, copies template files, and detects the intake path (SDR documents, wizard, or manual). Use when the user says 'new project', 'start a build', or provides intake documents.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/microsoft/MCS-Agent-Builder --skill mcs-init命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use this skill when the user asks Claude to autonomously test, fix, review, and merge a change end-to-end. Detects which lane(s) changed (frontend / backend / framework / agentspec), spawns relevant verifiers (agentic-test-loop with --require-oracle for false-green prevention; backend-verify for contracts/types/unit/server), runs a facilitator review loop using qa-challenger via Agent tool with worktree isolation until score>=9 and zero critical findings, calls multi-model-review.js review-merged as the final pre-merge oracle, and auto-merges via gh pr merge --auto if all gates pass and the change is not on the auto-merge denylist. User notified only on auto-merge completion or escalation. Triggered by /iterate, by phrases like 'make it work', 'iterate to green', 'fix and ship', or after Claude completes any non-trivial implementation in Layer A. Does NOT fire inside /mcs-* skills (they own their build/eval/fix loop) — for that, use /mcs-iterate.
Use this skill to build an agent in Copilot Studio from a researched agentspec.json. Runs pre-build validation (auth, env, connections, tools, model), then creates the agent, configures instructions, knowledge, tools, model, topics, and publishes. Supports single-agent, multi-agent, and connected-agent architectures. Use after /mcs-research when agentspec.json is ready, or to resume a partially completed build.
Use this skill to test an agent's quality after building or fixing it. Runs eval test sets against a published agent via Direct Line API (default, auto), CopilotStudio SDK, Power CAT Kit, or MCS Native Eval (for MCP agents). Results are written per-test to evalSets[].tests[].lastResult in agentspec.json. Use after /mcs-build or /mcs-fix, or when the user wants to re-test.
基于 SOC 职业分类
正在显示 SKILL.md
| name | mcs-init |
| description | Use this skill to start a new MCS agent project. Creates the folder structure in Build-Guides/, copies template files, and detects the intake path (SDR documents, wizard, or manual). Use when the user says 'new project', 'start a build', or provides intake documents. |
Create a new project folder and detect the intake path.
Provide project name:
/mcs-init ProjectNameBuild-Guides/[ProjectName]/
Check what's already in the folder:
Path A: SDR files found (.docx, .md, .pdf files that look like customer docs)
Found SDR files in Build-Guides/[ProjectName]/:
- [list files]
Next step: Run `/mcs-research ProjectName` to identify agents and research components
If .docx files exist, convert them to .md using pandoc:
pandoc "file.docx" -t gfm -o "file.md"
Note: pandoc is typically at %LOCALAPPDATA%\Pandoc\pandoc.exe — resolve via where pandoc or gcm pandoc.
Path B: No SDR files, start from scratch
No template needed — /mcs-research will create agentspec.json during Phase A (agent identification).
Guide the user to describe the agent or upload docs, then run research.
If SDR files detected:
## Project Initialized: [ProjectName]
**Location:** Build-Guides/[ProjectName]/
**SDR Files Found:**
- [list of converted .md files]
**Workflow:**
1. (Optional) Run `/mcs-context [CustomerName]` → Pull M365 history via WorkIQ
2. Click **Research** or run `/mcs-research [ProjectName]` → Read docs, identify agents, research components, enrich agentspec.json + generate evals
3. Click **Build** or run `/mcs-build [ProjectName] [agentId]` → Build in MCS
4. Click **Evaluate** or run `/mcs-eval [ProjectName] [agentId]` → Run tests
If starting from scratch:
## Project Initialized: [ProjectName]
**Location:** Build-Guides/[ProjectName]/
**Files Created:**
- (agentspec.json will be created when you run /mcs-research)
**Workflow:**
1. (Recommended) Run `/mcs-context [CustomerName]` → Pull M365 history via WorkIQ
2. Click **Research** or run `/mcs-research [ProjectName]` → Read docs, identify agents, research components, enrich agentspec.json + generate evals
3. Click **Build** or run `/mcs-build [ProjectName] [agentId]` → Build in MCS
4. Click **Evaluate** or run `/mcs-eval [ProjectName] [agentId]` → Run tests
Creates project folder, detects intake path, and guides user to next step.