Expose what the files do not answer
After this, you'll be able to create a source-backed question list and use it to make Claude's work with local files safer and easier to verify.
Before you start
Complete Compare versions without losing the thread first.
The idea
Good file work tells you what the files do not answer yet. This lesson asks you to make a source-backed question list, not a broad promise that Claude can use files. The output should be specific enough that you can inspect what Claude read, what it produced, and what still needs review.

Here is the before and after: Before, Claude fills gaps so the answer feels done. After, it lists questions tied to missing, conflicting, or weak source material. For example, a client folder might include notes, drafts, exports, and old versions. The lesson output should say which files are in scope, which are out, what Claude is making, and how you will prove the result came from the right source.
Now try it: Ask for questions before the final answer and decide which ones block the output. Make one choice before asking Claude to work: folder, file set, output shape, source check, access path, or maintenance rule. That choice keeps desktop file work from becoming vague local search.
The lesson is done when you can show the source boundary, the output, and the check that proves the output is safe to use.
Try it (15 min)
Watch out for
Paste this into Claude
Using this source inventory and task: [paste], create a source-backed question list. For each question, include Why It Matters, Source Gap, Blocks Final Output?, Who Can Answer, and Suggested Next Step. If any input is missing, ask me up to three questions before producing the artifact. Then return five sections: Finished Artifact, Realistic File Example, Assumptions To Check, What I Should Use In The Next Lesson, and One Risk If I Trust This Without Fixing It. Keep the answer practical enough that I can paste it into my folder workflow notes.
What a good response looks like
Finished Artifact: - a source-backed question list - Why it matters: it gives the folder workflow a concrete thing to inspect instead of a vague intention. - Use it next: paste this artifact into the next lesson before asking Claude to write, build, import, publish, or review anything. Reality Check: - The artifact names the user, input, decision, owner, or proof it depends on. - The weakest assumption is visible. - The next step can be completed in one sitting.
What good looks like
Go deeper (8 min)
Paste this into Claude
Without rereading the lesson, explain why a source-backed question list matters in three bullets. Then apply it to a second file example: [describe a different folder or document set]. Return What Changed, What Stayed The Same, What To Check Before Trusting The Output, and the exact next action.
What a good response looks like
Transfer Check: - What changed: the second example has a different audience, input, or delivery context. - What stayed the same: a source-backed question list still needs a source, a review check, and a next step. - Before trusting it: inspect the source check that would catch a wrong assumption. - Next action: run the check once, then carry the revised artifact into the next lesson.
What good looks like
When this breaks
AI can help with this
Paste the exercise prompt into Claude with your real folder or file context. Ask Claude to interview you one question at a time, produce a source-backed question list, challenge the weakest assumption, and rewrite the artifact once so it is ready for the next file step.

You can now
You can point to a source-backed question list.
Key takeaways
Source-backed questions make the final output more honest and easier to finish.
Go deeper