Microsoft Skill Creator

Microsoft skill creator

Create a Microsoft-technology skill by researching official Learn content, separating stable local knowledge from dynamic lookup guidance, and generating a portable skill package with clear triggers, references, and examples.

When to invoke

• "Create a GitHub Copilot skill for Azure Container Apps."
• "Build a skill that teaches agents our .NET SDK workflow."
• "Generate a Microsoft Learn backed skill for Semantic Kernel."
• "Make a skill for a Microsoft REST API with examples."
• "Create a Bicep or VS Code extension skill."

Prerequisites and context

• Prefer the Microsoft Learn MCP Server at https://learn.microsoft.com/api/mcp when available.
• If MCP tools are unavailable, use the mslearn CLI through npx @microsoft/learn-cli or a global mslearn install.
• Generate only portable skill package paths: SKILL.md, references/, sample_codes/, and assets/.
• Keep frontmatter minimal: name and description are required, and discovery depends primarily on the description.

Learn research tools

npx @microsoft/learn-cli search "semantic kernel overview"
npm install -g @microsoft/learn-cli
mslearn search "semantic kernel overview"

Generated skills should carry the same CLI fallback table so agents can continue work without the MCP server.

Procedure

1. Investigate the technology in three passes.
   • Scope discovery: microsoft_docs_search(query="{technology} overview what is"), microsoft_docs_search(query="{technology} concepts architecture"), and microsoft_docs_search(query="{technology} getting started tutorial").
   • Core content: microsoft_docs_fetch(url="...") for the best pages and microsoft_code_sample_search(query="{technology}", language="{lang}") for working code.
   • Depth: microsoft_docs_search(query="{technology} best practices") and microsoft_docs_search(query="{technology} troubleshooting errors").
2. Verify the investigation checklist: explain the technology in one paragraph, identify 3-5 key concepts, collect basic working code, name the common API patterns, and keep useful deeper-search queries.
3. Clarify with the user before generating: key areas found, primary agent tasks, and preferred programming language for samples.
4. Choose the template from references/skill-templates.md.
5. Generate the skill package and validate that local content is sufficient for common tasks, search queries return useful results, and code samples run.

Template selection

Local versus dynamic content

Store content locally when it is foundational, frequently accessed, stable, or hard to find. Keep content dynamic when it is exhaustive, version-specific, situational, or well indexed.

Investigation query patterns

Package shape

{skill-name}/
├── SKILL.md
├── references/
│   └── skill-templates.md
└── sample_codes/
    ├── getting-started/
    └── common-patterns/

For a Semantic Kernel skill, use semantic-kernel/, sample_codes/getting-started/hello-kernel.cs, sample_codes/common-patterns/chat-completion.cs, and sample_codes/common-patterns/function-calling.cs. Useful seed lookups include microsoft_docs_search(query="semantic kernel overview"), microsoft_docs_search(query="semantic kernel plugins functions"), microsoft_code_sample_search(query="semantic kernel", language="csharp"), microsoft_docs_fetch(url="https://learn.microsoft.com/semantic-kernel/overview/"), and microsoft_docs_fetch(url="https://learn.microsoft.com/en-us/semantic-kernel/frameworks/agent/agent-memory").

Progressive disclosure and bundled resources

• references/skill-templates.md: read when selecting the generated skill shape for SDK/Library, Azure Service, Framework/Platform, or API/Protocol work.

Compatibility terminology

Preserve these discovery and template terms when generating Microsoft skills: M365, NuGet/npm, classes/methods, end-to-end, general-purpose, getting-started/hello-kernel.cs, microsoft_docs_search(query="semantic kernel planner"), npm install -g @microsoft/learn-cli, npx @microsoft/learn-cli <command>, and PowerShell.

Open Horizons integration

• Scope new skills to the Developer IDP or Agent IDP objective and current Horizon stage.
• Preserve Open Horizons naming, routing, portability, package, and evidence boundaries.
• Route cross-domain sequencing through open-horizons-orchestration (skill).

Output template

## Microsoft skill package - <technology>

**Status:** generated | needs clarification | blocked
**Skill name:** `<skill-name>`
**Template:** SDK/Library | Azure Service | Framework/Platform | API/Protocol

### Research used
| Source | Query or URL | Purpose |
| --- | --- | --- |
| Microsoft Learn | `<query or URL>` | `<why it mattered>` |

### Local content included
- Core concepts: `<3-5 concepts>`
- Samples: `<sample paths>`
- Common patterns: `<patterns>`

### Dynamic lookup guidance
| Topic | Lookup |
| --- | --- |
| `<deeper topic>` | `microsoft_docs_search(query="...")` or `mslearn search "..."` |

### Validation
- Skill frontmatter: pass | fail
- Search queries tested: pass | fail
- Code samples checked: pass | fail

Quality gate

• name is kebab-case and matches the generated skill directory.
• description states what the skill does and when to use it with Microsoft technology triggers.
• The generated skill includes the Learn MCP to mslearn CLI fallback table.
• Stable essentials are local, while exhaustive or version-specific content is dynamic.
• references/skill-templates.md was used when choosing the template.
• Any referenced sample_codes/, references/, or assets/ paths exist.
• Code samples are working or explicitly marked as needing validation.

References

• Microsoft Learn MCP Server
• Semantic Kernel overview
• Semantic Kernel agent memory