# Skill catalog

Match the agreed architecture (root `README.md` in rapid mode, `./docs/spec.md` in SDD) against this catalog only when a companion capability is needed. `microsoft-foundry` is required. Use Copilot SDK for agents using runtime skills, and Microsoft Agent Framework for basic agents or explicit workflows. If a workflow also needs skills, clarify the framework choice. Explicit user framework choices override defaults; choose one framework per agent, not necessarily one for the entire solution.

Catalog membership is not proof that a skill still exists or is current. For a selected candidate, verify its source/version and available CLI support, using `gh skill preview <repository> <skill>` and installed skill metadata as appropriate. Respect pins and ask before installs or updates. Do not audit every installed skill or patch third-party copies. Project scope is the default installation scope unless the user chooses otherwise.

**Freshness preflight** (SDD Specify, or before relying on a selected skill's examples). Require GitHub CLI 2.90+. `gh skill list --json skillName,sourceURL,scope,version,pinned,path` proves installation, not freshness; compare tree SHAs with `gh skill update --dry-run`, and if a capability moved into a maintained parent or sub-skill, use that path. Mark each needed skill missing, current, update available or pinned. With approval, refresh unpinned stale skills with `gh skill update <skill>`; keep pins unless `--unpin` is approved. In read-only requests, report the command and the approval it needs without running it. When a skill conflicts with the installed package API, the package source wins; report the defect upstream as an issue or PR instead of patching the local copy. Foundry roles come from the maintained `microsoft-foundry/rbac` sub-skill, not standalone `azure-rbac`.

| Repository                | Skill                          | Suggest when the architecture needs |
| ------------------------- | ------------------------------ | ------------------------------ |
| `microsoft/azure-skills`  | `microsoft-foundry`            | Required platform executor for Agentic Loop; use its maintained RBAC guidance for Foundry management-plane and data-plane permissions |
| `microsoft/azure-skills`  | `azure-ai`                     | AI Search, vector/hybrid search, semantic search, Document Intelligence/OCR, Speech/STT/TTS |
| `microsoft/azure-skills`  | `azure-aigateway`              | AI Gateway, API Management for model/tool/agent routing, semantic caching, content safety, token limits, MCP governance |
| `microsoft/azure-skills`  | `appinsights-instrumentation`  | Application Insights SDK setup, telemetry instrumentation, traces, metrics, and app monitoring |
| `microsoft/azure-skills`  | `azure-kusto`                  | KQL analytics, Azure Data Explorer, Log Analytics-style telemetry exploration, time-series analysis |
| `microsoft/azure-skills`  | `azure-storage`                | Blob/File/Data Lake storage for grounding data, uploads, durable artifacts, or document/file landing zones |
| `microsoft/azure-skills`  | `azure-messaging`              | Service Bus, Event Hubs, async business actions, event ingestion, queues, topics, or messaging SDK troubleshooting |
| `microsoft/azure-skills`  | `entra-app-registration`       | Microsoft Graph/OAuth app registration, delegated auth, API permissions, MSAL integration |
| `microsoft/azure-skills`  | `entra-agent-id`               | Entra Agent Identity, agent OAuth, OBO, workload identity federation, cross-tenant agent auth |
| `microsoft/azure-skills`  | `azure-prepare`                | SDD services beyond the azd AI starter, or explicitly chosen IaC/azd scaffolding in rapid mode |
| `microsoft/azure-skills`  | `azure-validate`               | Validation of an azd/IaC deployment, including configuration, RBAC and what-if checks |
| `microsoft/azure-skills`  | `azure-deploy`                 | Execution/recovery of a prepared azd/IaC deployment |
| `github/awesome-copilot`  | `microsoft-agent-framework`    | Basic agents or explicit agent workflows, or an explicit MAF choice; clarify workflows that also need runtime skills |
| `github/awesome-copilot`  | `copilot-sdk`                  | Agents using runtime skills, or an explicit Copilot SDK choice; BYOK Foundry models, toolbox tools/skills and the Responses API |
| `github/awesome-copilot`  | `python-mcp-server-generator`  | A Python MCP server (FastMCP, streamable HTTP) |

Build skills help the coding assistant build the app; [run skills](run-skills-catalog.md) are executed by the resulting agent. Follow current `microsoft-foundry` workflows for governed skill versions and toolbox configuration, and verify generated code against installed SDK APIs. Catalog selection must not override the mode defaults in `SKILL.md`. This skill, the canvas and the workflow ship with the `foundry-agentic-loop` plugin; do not install them as companions.