Audit Claude's Context of You and Your Work in 15 Minutes

Start by asking Claude what it thinks it knows before you correct anything. This is important because the audit only works if Claude separates what it knows from what it is guessing.

Use this prompt:

Audit your context and memory assumptions about me.

Put them in a table with:
- what you believe
- why you believe it
- your confidence level
- whether each item is confirmed, an assumption, outdated, or unknown

Cover:
- my role
- priorities
- KPIs
- tools
- workflows
- anything you may be over-weighting from old chats or side projects

Do not interview me yet.

Claude should return a table that shows what it believes about your work. Some of it may be correct. Some of it may be stale. Some of it may be random context from a side project that should not affect your real work.

In the recording, Claude correctly understood some core responsibilities, but it also over-weighted one-off projects, personal questions, old trainings, and tools that were not actually part of the main job.

That is the point. You are not trying to prove Claude is wrong. You are trying to see what context it is using.

Pro tip: Run this audit anywhere you do serious AI work. Different tools may have different assumptions about you.

Once Claude shows the table, do not try to correct everything manually. Ask it to turn the uncertain items into an interview.

Use this prompt:

Now interview me about the assumptions, outdated items, and unknowns from that audit.

Ask in rounds.
Use multiple-choice questions wherever possible.
When multiple choice is not enough, ask for one or two sentences.
Cover my role, KPIs, tools, priorities, and work processes.
After each round, summarize what changed.

The multiple-choice part matters. Without it, this can become a long homework assignment. With it, Claude can ask questions that are fast to answer.

For example, it may ask whether a project is still active, whether a workflow is part of your real job, or whether something was just a one-off test. You can usually answer with a number, a letter, or one sentence.

Pro tip: The first two prompts are usually enough. Claude can often walk through the interview rounds on its own after that.

Claude should walk through several rounds. In the recorded demo, it moved through:

1. role, mission, and main responsibilities 2. current priorities and what is top of mind 3. follow-up questions about those priorities 4. tools, systems, and where the friction really lives 5. success signals and what "good" looks like

Do not worry about answering perfectly. The goal is to correct the big picture.

Look for places where Claude thinks a side quest is core work. For example, if you built a random website for a demo, Claude might treat that as an active project. If you asked about travel, personal research, or a one-off tool test, Claude might keep that context around longer than you want.

This is where you tell Claude what matters now.

If Claude asks an open-ended question, keep the answer short. A sentence or two is enough. The audit should make your work context clearer, not create another huge document you have to maintain.

After the interview, Claude may offer to update its memory. If it does, review the proposed updates and approve the ones that are accurate.

If Claude does not offer, ask directly:

Update your memory based on the corrected context from this audit.

This is where the workflow becomes more than a report. You are helping Claude clean up stale assumptions and replace them with a more current picture of your role, responsibilities, priorities, tools, workflows, and success signals.

Use judgment here. Do not save sensitive data, private numbers, credentials, client details, or anything you would not want available as general assistant context. Save stable work context only.

Once Claude has the updated context, ask it to create a report. This is the durable artifact from the audit.

Use this prompt:

Create a Markdown artifact with a report of the job we completed.

Include:
- what we audited
- what changed in your understanding
- recommendations for automations and skills I should build
- anything I should cut to improve my work process
- any context that should be reviewed again later

In the recording, Claude created a report with the updated work context, headline findings, bottlenecks, skill ideas, automation ideas, and things to cut.

That is more useful than a generic "about me" file. It gives you a snapshot of how you work right now, where your workflow is noisy, and what Claude thinks would make the biggest difference.

Save that Markdown file somewhere you can find it later. You can also use it as the source for an about-me.md file, a Claude Project instruction, or a team operating doc.

Pro tip: Ask for things to cut, not just things to build. The best recommendation may be removing a workflow Claude keeps over-weighting.

The report should not just sit there. Read it like an operations review.

Look for three categories:

• Automations to build: repeated work Claude noticed during the interview.
• Skills or prompts to create: workflows that would be easier if they were repeatable.
• Things to cut: projects, habits, tools, or context that distract from your real priorities.

This is where the audit becomes useful for the user, not just the AI. Claude may identify bottlenecks you have been working around for weeks. It may also point out that some "projects" are really old experiments that should stop influencing your priorities.

The best outcome is not that Claude knows every detail about you. The best outcome is that Claude knows what to optimize for.

Ask Claude to turn the audit into a reusable skill:

Turn this interview and report-generation process into a reusable skill.

The skill should:
- audit your current assumptions about my work
- interview me with multiple-choice questions where possible
- update your context based on my answers
- generate a Markdown report with recommendations
- suggest automations, skills, and things to cut

Then rerun the audit every quarter.

Your role, KPIs, tools, and priorities change over time. Your AI context should change with them.