Your context files go stale. Priorities change, the work moves on, and the files quietly stop describing you — but the output only gets slightly worse, so nobody notices. This prompt makes Claude tell you exactly what it still doesn't know about you, and what to add next.
This prompt reads your context files and grades them. If you haven't created them yet, it has nothing to work with.
Build them first — a skill interviews you and writes all three in about five minutes.
Get the Starter Kit →A one-time install. After that, say "audit my context" in any Cowork project and it runs — and unlike the prompt, the skill can go and look at your actual work first.
It reads your context files, optionally gathers evidence from your real work, then names the gaps, scores each file out of 5, and writes the fixes.
⬇ Download context-audit.skillCustomize → Skills → + and drop the file in.A prompt can only grade what your files say about themselves. The skill can ask permission to look at your real work first — and then the audit gets much sharper, because it can catch the gap between what you wrote down and what you actually do.
It might come back with: "your priorities list says brand campaigns, but everything in this folder for two months has been localisation" — or "your voice file says you avoid jargon, but here are three lines where you used it." That's the finding you can't get from a prompt.
No install, works anywhere. This version audits your files only — it won't go looking at your work. Open the project, paste, send.
Any Claude · reads whatever files are thereAudit the context you have about me. Be blunt — I want the gaps, not reassurance. STEP 1 — Read every file in this project folder, especially about-me.md, personal-voice.md and working-style.md, plus any global instructions you have. STEP 2 — Tell me, in plain language, what you actually know about me right now: my role, who I work for, what I'm prioritising, how I sound, and how I want to be worked with. Two short paragraphs. If any of it is inferred rather than written down, say which. STEP 3 — Now the useful part. Name the gaps, specifically. For each one: what's missing, which file it belongs in, and why it would change your output. Cover at least: - Facts about my work you'd have to guess at - Anything that reads like it's out of date - Voice: could you write a paragraph that I'd believe I wrote? What's missing to get there — more samples, sharper rules, words to avoid? - Working style: is there anything you still have to ask me every time? - Contradictions between files, or between the files and how I've actually behaved in this project STEP 4 — Score each of the three files out of 5 for how usable it is, with one sentence saying why. Don't be generous. STEP 5 — Give me the fixes as ready-to-paste text. For each gap, write the actual lines to add and tell me which file to put them in. Don't describe the change — write it. STEP 6 — Then ask me the three questions that would close the biggest gaps, one at a time. When I answer, update the files directly. Be specific and honest. If a file is thin, say so plainly — a vague audit is worse than none.
Setting up context is a one-off task that everyone does once and never revisits. But the files are the thing doing the work — the model didn't get better at sounding like you, your files did. Auditing them every month or two is the cheapest quality improvement available to you, and it takes about ten minutes.
Haven't built the files yet? Start with the Cowork Starter Kit — a skill interviews you and writes all three.