Microsoft Technical Research

Deep technical research on Microsoft technologies with multi-agent orchestration and structured documentation output.

Skills

Skill	Description
initiate-microsoft-researcher	MANDATORY entry point — validates topic, confirms plan with user, scaffolds folder structure, creates `.initiated` marker.
create-microsoft-research-plan	Generate structured research plans with scope, areas, search terms, and output mapping.
research-note-template	Extract structured notes from sources using standardized extraction rules.
research-output-scaffolding	Scaffold the .research/ folder structure with note directories and section placeholders.

Agents

Agent	Description
research-microsoft-technology	Orchestrator — coordinates the full research lifecycle from plan to finished output.
research-source-discovery	Discovers and ranks candidate sources across all enabled research areas.
research-deep-reader	Fetches sources and extracts structured notes following extraction rules.
research-content-writer	Synthesizes notes into polished output sections with full attribution.
research-quality-reviewer	Reviews output for attribution, completeness, and source validity.

Research Workflow

┌────────────────────────────────────────────────────────────────────┐
│ Phase 1 — INITIATE (mandatory entry point)                        │
│                                                                    │
│  ┌─────────────┐    ┌──────────────┐    ┌──────────────────┐      │
│  │  Plan Skill │───▶│ User Confirm │───▶│   Scaffolding    │      │
│  └─────────────┘    │  (BLOCKING)  │    └──────────────────┘      │
│                     └──────────────┘           │                  │
│                                          .initiated ✅            │
└────────────────────────────────────────────────┼───────────────────┘
                                                 │
                                                 ▼
                                          ┌─────────────┐
                                          │ Orchestrator │
                                          └──────┬──────┘
                                                 │
                  ┌──────────────────────────────┼────────────────────┐
                  ▼                              ▼                    ▼
         ┌────────────────┐          ┌────────────────┐   ┌──────────────────┐
 Phase 2 │Source Discovery│ Phase 3  │  Deep Reader   │ 4 │ Content Writer   │
         └────────────────┘          └────────────────┘   └──────────────────┘
                                                                    │
                                                                    ▼
                                                         ┌──────────────────┐
                                                  Phase 5│Quality Reviewer  │
                                                         └──────────────────┘
                                                                    │
                                                            Phase 6 ▼
                                                         ┌──────────────────┐
                                                         │ Iterate/Complete │
                                                         └──────────────────┘

Important: The orchestrator agent is FORBIDDEN from performing research directly. It MUST dispatch subagents for all research phases. The initiate-microsoft-researcher skill is the only entry point and creates a .initiated marker that the orchestrator checks before dispatching any subagent.

Tool Restrictions

The orchestrator agent intentionally has a restricted tool set (vscode, read, agent, edit, todo) — it cannot access web, browser, search, or microsoft-learn/* tools. This is enforced at the platform level to prevent the orchestrator from bypassing subagents and performing research directly. Each subagent has only the tools it needs (principle of least privilege).

Progress Reporting

The orchestrator reports a brief status update to the user between every phase (e.g., "Phase 2 complete — discovered 23 sources across 5 areas. Moving to Phase 3…"). This gives visibility during autonomous research phases.

Output Structure

All research output is written to .research/<topic-slug>/ in the workspace root:

.research/<topic-slug>/
├── plan.md              # Research plan (scope, areas, search terms)
├── log.md               # Activity log (all agent actions)
├── sources.md           # Discovered sources with relevance scores
├── review.md            # Quality review report
├── notes/
│   ├── docs/            # Microsoft Learn documentation notes
│   ├── tech/            # Technical documentation notes
│   ├── blogs/           # Blog and article notes
│   ├── arch/            # Architecture pattern notes
│   ├── samples/         # Code sample notes
│   ├── solutions/       # Solution accelerator notes
│   └── other/           # Community and third-party notes
└── output/
    ├── README.md        # Section index with status
    ├── 01-overview.md   # Synthesized output sections
    ├── 02-architecture.md
    └── ...

Research Purposes

Purpose	Description	Target Audience
`deep-guide`	Comprehensive technical reference	Architects, senior engineers
`presentation`	Tech talk or demo deck	Conference/meetup attendees
`lab`	Hands-on workshop exercises	Developers learning through practice
`demo`	Sample solution or proof-of-concept	Developers evaluating for adoption

Usage

VS Code (Agent Mode)

Select the research-microsoft-technology agent in the Copilot Chat panel, then describe the research topic and purpose:

@research-microsoft-technology research Azure Container Apps dynamic sessions for a deep-guide

Or switch to the agent mode and invoke it naturally:

Research Azure AI Foundry agent evaluation for a presentation

The agent will invoke the initiate-microsoft-researcher skill, present the plan for your approval, and then coordinate all subagents automatically.

Copilot CLI

Use the --agent flag to invoke the orchestrator from the terminal:

gh copilot chat --agent research-microsoft-technology \
  "Research Azure Container Apps dynamic sessions for a deep-guide"

Important Notes

Always invoke the research-microsoft-technology agent — do not invoke the initiate-microsoft-researcher skill directly. The skill is non-user-invocable and is called internally by the orchestrator agent.
The orchestrator always begins with the initiate-microsoft-researcher skill. You will be asked to confirm the plan before any research starts.
If you provide source URLs in your prompt, they will be included in the plan's search terms automatically.
All output is written to .research/<topic-slug>/ in the workspace root.
You can resume an interrupted research project by invoking the agent again with the same topic — it will detect the existing .initiated marker and continue from where it left off.