microsoft/hve-core
Publicmirrored fromhttps://github.com/microsoft/hve-coreAvailable
docs/getting-started/first-research.md
79lines · modecode
| 1 | --- |
| 2 | title: Your First Research |
| 3 | description: Use the task-researcher agent to investigate your own codebase |
| 4 | sidebar_position: 5 |
| 5 | author: Microsoft |
| 6 | ms.date: 2026-02-18 |
| 7 | ms.topic: tutorial |
| 8 | keywords: |
| 9 | - getting started |
| 10 | - first research |
| 11 | - task researcher |
| 12 | - rpi workflow |
| 13 | - github copilot |
| 14 | estimated_reading_time: 5 |
| 15 | --- |
| 16 | |
| 17 | > [!NOTE] |
| 18 | > Step 2 of 4 in the [Getting Started Journey](./). |
| 19 | |
| 20 | The RPI framework separates research from implementation. Before writing code, |
| 21 | you find verified facts about the codebase. This exercise introduces the |
| 22 | Research phase by itself, without the planning or implementation phases. |
| 23 | |
| 24 | ## Pick a Research Question |
| 25 | |
| 26 | Choose something you genuinely want to know about your codebase. Examples: |
| 27 | |
| 28 | * "How are tests structured in this project?" |
| 29 | * "What patterns does this codebase use for error handling?" |
| 30 | * "What dependencies does the authentication module have?" |
| 31 | |
| 32 | Use a question where the answer would actually help you. Contrived exercises |
| 33 | teach methodology; real questions teach methodology and produce useful output. |
| 34 | |
| 35 | ## Run the Research |
| 36 | |
| 37 | 1. Open GitHub Copilot Chat (`Ctrl+Alt+I`). |
| 38 | 2. Select the **task-researcher** agent. |
| 39 | 3. Type your question as a prompt. |
| 40 | |
| 41 | The agent creates a research document in `.copilot-tracking/research/` with |
| 42 | findings, file references, and evidence. This takes 2-5 minutes depending on |
| 43 | the scope of the question. |
| 44 | |
| 45 | ## Read the Output |
| 46 | |
| 47 | Open the research document. You'll find: |
| 48 | |
| 49 | * The agent cites file references with line numbers for each fact. |
| 50 | * Conclusions trace back to actual code through evidence-linked findings. |
| 51 | * Areas where research was incomplete appear as remaining questions. |
| 52 | |
| 53 | This is what "verified truth" looks like in RPI. The agent did not guess or |
| 54 | generate plausible answers. It searched, read, and cited. |
| 55 | |
| 56 | > [!NOTE] |
| 57 | > Your first research document may feel verbose. That's intentional. Research |
| 58 | > outputs are reference material for the planning phase, not finished prose. |
| 59 | > Over time, you'll learn to scope questions tightly to get focused results. |
| 60 | |
| 61 | ## What You Learned |
| 62 | |
| 63 | * Researching before implementation reduces guesswork and rework. |
| 64 | * Agents produce artifacts like research documents, not chat messages alone. |
| 65 | * Artifacts carry context so the next phase builds on verified facts rather |
| 66 | than starting from scratch. |
| 67 | |
| 68 | ## Next Step |
| 69 | |
| 70 | You've installed HVE Core, talked to an agent, and run your first research. |
| 71 | The next step is a full Research, Plan, Implement cycle: |
| 72 | [Your First Full Workflow](first-workflow.md). |
| 73 | |
| 74 | --- |
| 75 | |
| 76 | <!-- markdownlint-disable MD036 --> |
| 77 | *🤖 Crafted with precision by ✨Copilot following brilliant human instruction, |
| 78 | then carefully refined by our team of discerning human reviewers.* |
| 79 | <!-- markdownlint-enable MD036 --> |
| 80 | |