Context Cleanup

Repair the information around a task before asking a tool to act

What you will make

Context is not automatically useful because it is available. Learners practice finding the current source of truth, distinguishing stale or untrusted material from useful support, and making an explicit cleanup decision before any downstream work begins.

By the end, you can identify missing, stale, conflicting, excessive, and untrusted context; explain what should be added, corrected, removed, or isolated; and produce a concise, reviewable cleanup brief.

Lab materials

Use during the lab

• Documents
  • Fictional project packet: neighborhood learning session Required · Markdown · A fictional packet containing current, stale, conflicting, informal, and private context.
• • Lab Notebook Interactive · An optional place to record the work and download a local copy.
• • Learner guide Markdown

Go deeper

• Related articles
  • Before You Ask AI for Help, Clean Up the Context Article
• • Facilitator guide Markdown

Before you start

Level and format: Beginner; individual, pair, or small group

You need: the reusable project packet, the optional context cleanup worksheet or another writing surface. A chat tool is optional for the final restatement check. Use invented or safely redacted material; do not put private client, employee, financial, health, legal, or other consequential information into a consumer tool.

The core idea

The first useful action is often cleanup, not another request. A source hierarchy and a change log make uncertainty visible instead of allowing an AI tool to smooth it into plausible prose.

Clean the context before the work

Complete the activity here. Open the optional Lab Notebook beside this page when you want a place to record your work.

1. Read the packet

   Work from the fictional project packet, or use another safely redacted collection that contains a current source, old material, a conflict, an informal note, and something that should not be shared.

2. Name the source of truth

   Identify the current authoritative source and explain why it outranks the other material. If no source is authoritative, name that uncertainty rather than guessing.

3. Sort the context

   Mark each item as missing, stale, conflicting, excessive, or untrusted. One item may belong in more than one category; record why.

4. Make the cleanup decisions

   For each item, choose: add, correct, remove, isolate, retain for reference, or verify with a responsible person or source. Keep a short change log.

5. Create a clean context brief

   Include the task, audience, current facts, active constraints, source hierarchy, excluded material, and unresolved questions. This is not the final deliverable; it is the information a later task may safely use.

6. Run a restatement check

   If you use a chat tool, ask: “Restate the goal, source of truth, active constraints, excluded material, and unresolved questions. Flag conflicts before doing any work.” Correct the summary if it makes an unsupported assumption.

7. Record remaining uncertainty

   Name what still needs verification and who owns that decision. Stop rather than filling a gap with an invented answer.

Check your result

Ask: What could go wrong if a later tool treated this item as current? Did we remove something merely inconvenient, or something irrelevant or unsafe? What is the difference between an authoritative source and a plausible note? What uncertainty remains after cleanup?

You can compare the original packet, change log, and clean brief. You cannot prove that a tool will always honor a hierarchy, or that a brief contains every relevant fact. For real work, a responsible person still verifies dates, policies, commitments, and access permissions.

• Which item looked useful but should not have been included?
• What did you treat as authoritative, and why?
• Which cleanup decision mattered most for a later task?
• What would you need to verify before using the clean brief in real work?

A polished result is not proof that it is accurate, complete, safe, or appropriate. Keep unresolved questions visible and verify consequential claims outside the tool.

You are done when

Save the annotated packet, source hierarchy, cleanup change log, clean context brief, and a short reflection. Done means you can identify one source of truth, explain at least three cleanup decisions, and name unresolved uncertainty without asking a tool to invent an answer.

• The low-risk task or supplied packet
• The actual goal and intended audience
• A visible sort of context to keep, remove, correct, or question
• A cleaned context brief
• A comparison showing what changed
• One item requiring human review or outside verification
• One thing I deliberately did not share

Try another version

These are optional extensions, not additional completion requirements.

• Restore one removed detail and observe whether it helps or distracts
• Ask a partner to review the context sort before seeing the output
• Create a shorter brief without removing a required constraint
• Repeat with another tool and record variation without declaring one universally best

Learner Map

Tracks

• AI Literacy for the Rest of Us

Modules

• Context Engineering

Help us improve this Activity

Tell us what helped, what slowed you down, or what you needed but could not find.

🎓 Give learner feedback