Skip to content
GitHub Copilot Train-the-Trainer
Lab

Session M7-01 lab: Build and check a grounded working brief

Duration: 2 hours

Difficulty: Beginner

Deliverable: initiative-context-brief.md, ready for Session M7-02 without the original chat

What you will learn

Compare Copilot's answers with and without project context. Use the sourced information to write a brief, then revise it when new information arrives. Finally, check whether a fresh session can use the brief on its own.

Part Work Time
1 Check the app, mode, and project 15 min
2 Compare weak and grounded context 25 min
3 Create the first working brief 30 min
4 Update the brief with new source information 25 min
5 Check the brief in a fresh session 25 min

Before you start

Complete the track capability setup.

You need:

  • GitHub Copilot app access;
  • read access to the training repository project;
  • permission to create a branch or working copy for the brief.

Check GitHub Copilot access before starting. If GitHub Copilot access or project access is unavailable, stop. Do not continue in a generic chatbot.

Part 1: Check the app and project (15 minutes)

  1. Open the training repository under Projects.
  2. Start an interactive session.
  3. Select Interactive mode. Type @ and add lab/starter/initiative-overview.md.
  4. Ask Copilot to report the file name and its first heading.
  5. Open Customize → Installed and inspect the visible skills, MCP servers, plugins, and canvases.
  6. Run /context.

Use Interactive for the comparison. Switch to Plan before asking Copilot to propose repository changes. Do not use Autopilot for this lab because you need to review each write.

@ adds files, # references issues, and / opens commands in the app. Type the character and select the item from the picker. A typed file token alone does not show that the app attached the source.

Checkpoint: Copilot can read the supplied file, and you can identify which capabilities are available.

Part 2: Compare weak and grounded context (25 minutes)

Start a fresh session with /clear or /reset so the file read in Part 1 is not carried into the comparison. Do not attach the initiative file yet. Start with this deliberately weak prompt:

Explain the Service Request Portal initiative, its users, constraints, and next
decision. Mark anything uncertain.

Identify statements that have no visible source.

Now attach starter/initiative-overview.md and ask:

Use only @initiative-overview.md. Explain the requested outcome, affected users,
known constraints, and next decision. Cite the heading that supports each fact.
Keep missing information as unknown.

Compare the answers:

Check Unscoped answer Grounded answer
Unsupported facts
Visible unknowns
Source references
Useful next decision

Checkpoint: Explain which answer is safer and how repository context changed it.

Part 3: Create the first working brief (30 minutes)

Copy the templates:

cp sessions/session-m7-01-copilot-app-foundations/lab/starter/initiative-context-brief-template.md \
  docs/discovery/initiative-context-brief.md

Ask Copilot to complete the brief from initiative-overview.md. Require:

  • the requested outcome and affected users;
  • known facts with section references;
  • assumptions separated from unknowns;
  • named non-goals;
  • the next decision and its owner;
  • the relevant capabilities found in Part 1.

Review the proposed change before accepting it. Reject any statement that has no source, invents an owner, or turns an unknown into a fact.

Commit the first version:

git add docs/discovery/initiative-context-brief.md
git commit -m "Create grounded initiative working brief"

Checkpoint: Your brief separates source facts, assumptions, unknowns, and the next decision clearly enough for someone to use it without reading the chat.

Part 4: Update the brief with new source information (25 minutes)

Open starter/late-evidence.md. Treat it as source information the owner approved after the first brief was committed.

Ask Copilot to:

  1. compare the new information with the saved brief;
  2. list which sections need to change;
  3. identify any resolved unknown;
  4. identify any statement that is now contradicted;
  5. propose a focused file update.

Review the diff. Do not rewrite unrelated sections.

Commit the revision:

git add docs/discovery/initiative-context-brief.md
git commit -m "Update working brief with approved evidence"

Checkpoint: Git history shows which source changes led to the revision.

Part 5: Check the brief in a fresh session (25 minutes)

Before starting fresh, run /context. If the current session is crowded but the thread still matters, use /compact. For the fresh-session check, use /clear or /reset, then attach only docs/discovery/initiative-context-brief.md.

Ask:

Using only this working brief, report:
1. the requested outcome;
2. supported facts and their sources;
3. unresolved decisions;
4. named non-goals;
5. the next owner action.
Do not infer missing information.

Compare the answer with the file. If the answer misses or misstates any of these items, correct the brief so the new session has what it needs.

Run an independent critique:

/rubber-duck Review the initiative context brief for unsupported claims,
ambiguous ownership, hidden customer references, and decisions presented as facts.

If /rubber-duck is unavailable, stop this step. The trainer must check the current documented critique option before delivery rather than inventing a replacement command.

Final deliverable

Submit:

  1. docs/discovery/initiative-context-brief.md;
  2. two commits showing the first brief and the revision based on the new source.

Session M7-02 uses this brief as the source for its decision interview.

Verification

  • Copilot read the repository project successfully.
  • The unscoped and grounded answers were compared.
  • Every fact in the brief points to a source section.
  • Assumptions and unknowns remain separate.
  • New source information produced a focused revision.
  • A fresh session identified the outcome, facts, unresolved decisions, non-goals, and next owner action from the brief alone.
  • No customer or source organization names appear.