Initiate microsoft researcher

Initiate Microsoft Researcher

Mandatory entry point for all research orchestration, called exclusively by the research-microsoft-technology orchestrator agent. This skill ensures the research plan is created, explicitly confirmed by the user, and the folder structure is scaffolded before any research subagents are dispatched.

This skill is the ONLY way to begin a research project. No exceptions.

Why This Exists

Without a formal initiation gate, the orchestrator agent may bypass the structured workflow — fetching sources directly, writing notes inline, or skipping subagent coordination entirely. This skill enforces the contract between the orchestrator and the user.

Prerequisites

• A clear technology topic from the user
• A defined purpose (deep-guide, presentation, lab, or demo)

Process

Step 1 — Collect Topic and Purpose

If the user has not provided both a topic and purpose, use vscode_askQuestions to collect:

1. Topic: The Microsoft technology, feature, architecture, or solution
2. Purpose: What the research is for (deep-guide, presentation, lab, or demo)
3. Source URLs (optional): Any specific URLs the user wants included in the research
4. Additional context (optional): Audience, event name, constraints, deadlines, or special requirements

Derive the kebab-case topic slug from the topic name.

Step 2 — Generate the Research Plan

Invoke the create-microsoft-research-plan skill with:

• Topic and topic slug
• Purpose
• Any user-provided source URLs (to include in the plan's search terms)
• Additional context

This produces .research/<topic-slug>/plan.md.

Step 3 — Present Plan for User Confirmation (BLOCKING)

This step is a hard gate. Research CANNOT proceed without explicit user approval.

Present the plan to the user showing:

1. Topic slug and folder path
2. Purpose and target audience
3. Enabled research areas with search terms
4. Expected output sections
5. Estimated source count targets per area
6. Any user-provided source URLs included

Use vscode_askQuestions with the following options:

• Approve: Plan is correct, proceed to scaffolding
• Revise: User provides feedback, return to Step 2 with adjustments
• Cancel: Abort the research project entirely

If the user selects "Revise": Apply their feedback, regenerate the plan, and present again. Repeat until approved or cancelled.

If the user selects "Cancel": Stop immediately. Do not scaffold or dispatch subagents.

If the user selects "Approve": Proceed to Step 4.

Step 4 — Scaffold the Folder Structure

Invoke the research-output-scaffolding skill with:

• Topic slug
• Purpose
• Enabled areas from the approved plan

This creates the full .research/<topic-slug>/ directory tree including notes/, output/, log.md, and section placeholders.

Step 5 — Initialize the Research Log

Verify the log file exists at .research/<topic-slug>/log.md and append the initiation entry:

- [{{TIMESTAMP}}] INITIATE: Research project started
  - Topic: {{TOPIC}}
  - Purpose: {{PURPOSE}}
  - Plan: APPROVED by user
  - Areas: {{ENABLED_AREAS}}
  - Output sections: {{SECTION_COUNT}}
  - User-provided sources: {{SOURCE_COUNT}} URLs

Step 6 — Create Initiation Marker

Create .research/<topic-slug>/.initiated with:

initiated: true
approved_by: user
approved_at: {{TIMESTAMP}}
topic: {{TOPIC}}
purpose: {{PURPOSE}}
plan_hash: {{MD5_OF_PLAN}}

This marker file is checked by the orchestrator agent to verify that initiation was completed before dispatching subagents.

Step 7 — Return Control to Orchestrator

Return a structured summary to the orchestrator:

## Research Project Initiated

- **Topic**: {{TOPIC}}
- **Slug**: {{TOPIC_SLUG}}
- **Purpose**: {{PURPOSE}}
- **Plan**: Approved ✅
- **Folder**: `.research/{{TOPIC_SLUG}}/`
- **Areas**: {{ENABLED_AREAS}}
- **Output sections**: {{SECTION_IDS}}
- **Ready for**: Phase 2 — Source Discovery

The orchestrator MUST now dispatch `research-source-discovery` as a
subagent. Do NOT fetch sources directly. Do NOT write notes directly.
Do NOT skip any phase.

Constraints

• NEVER skip user confirmation — the plan must be explicitly approved
• NEVER proceed to source discovery or deep reading from this skill
• NEVER fetch source content — that is the deep-reader's responsibility
• NEVER write research notes — that is the deep-reader's responsibility
• NEVER synthesize content — that is the content-writer's responsibility
• This skill only creates the plan, confirms it, and scaffolds the structure
• If vscode_askQuestions is unavailable, present the plan in chat and ask for explicit textual confirmation before proceeding