The creator-seat audit: 4 Claude prompts

You are helping me run a creator-seat audit on my own work. My job is [job title] at [type of company], and here is an honest list of everything I actually did in a normal week, with rough time spent and how often each one recurs: [paste your list].

Sort every item into exactly one of four buckets and show it as a table with columns for task, hours per week, bucket, and a one-line reason. The buckets are: (1) pure repetition, where the same steps produce the same shape of output every time, (2) repetition with judgment inside, where the steps repeat but a human call is buried in the middle, (3) judgment, where the value is the decision rather than the output, and (4) creation, where I am defining something that did not exist before. Be blunt with me. If I have flattered a task by calling it strategic when it is really a template, say so and say why.

Then total the hours in each bucket and tell me what percentage of my week is currently spent in buckets one and two. Rank the bucket-one and bucket-two tasks by how much time I would get back per month if they were automated, and for the top three, describe in plain language what a good automated version would need to know about my format, my tone, and my inputs to produce work I would actually sign off on.

Finish with the honest part: name the two things in my week that are genuinely hard to replicate and explain what specifically makes them hard, whether that is taste, relationships, context nobody wrote down, or accountability for the outcome. Do not reassure me. If very little of my week sits in buckets three and four, tell me that plainly and tell me which single task is the most realistic bridge into creation work.

Build me a reusable Claude skill for the task we ranked number one in my audit: [name the task, for example "the Monday client status report"]. I am not a developer, so write the whole thing for me and explain nothing in code terms.

Before you write it, ask me the four or five questions you genuinely need answered to get this right: what the inputs look like, what the finished output has to look like, what tone or house style it must match, what it must never do, and where a human decision has to stay with me. Wait for my answers.

Then write the SKILL.md. Give it a short lowercase hyphenated name that does not contain the words claude or anthropic, a description line that says both what it does and exactly when it should trigger, and numbered instructions specific enough that a competent stranger could follow them and produce the same result I would. Where my audit said a judgment call sits inside this task, write an explicit instruction to pause and ask me rather than guessing.

Finally, tell me how to save it as a zipped folder I can upload, and give me one realistic test case I can run on it this week to check whether the output is genuinely sign-off ready or only nearly there.

I have freed up roughly [number] hours a week by automating repetitive parts of my job at [type of company], where I work as [job title]. I want to spend those hours on creation work: building something that did not exist before, that only I am well placed to build, and that makes my judgment visible to people who decide things.

First, ask me three questions about what my team or company is currently bad at, what problem I complain about most, and who would notice if it were fixed. Wait for my answers.

Then give me five candidate projects, each described in four lines: what it is, who it helps, why I am unusually well placed to build it given my role and access, and what the first visible version could look like within two weeks. Rank them by a simple test: how obvious would the value be to someone two levels above me who only sees the finished thing. Reject anything that requires permission I do not have or a budget I have to ask for.

For the top one, write me a two-week plan broken into six sessions of ninety minutes, each with a specific deliverable, and tell me the single most likely reason I will abandon it in week one and what to do instead when that happens.

We ran a creator-seat audit on my work previously. Here is what my week looks like now: [paste your current recurring tasks with rough hours]. Here are the skills or automations I already have running: [list them].

Re-sort my week into the same four buckets: pure repetition, repetition with judgment inside, judgment, and creation. Compare the percentages to last time if I have given you the earlier numbers, and tell me plainly whether I moved toward creation work or away from it.

Then do three things. Name any new repetitive task that has crept in since last time and is now worth automating. Name any existing skill of mine that is probably underperforming, based on tasks I am clearly still doing by hand, and tell me what to fix in its instructions. And name one thing I am still doing manually out of habit rather than necessity, where the real reason is that handing it over feels uncomfortable rather than risky.

End with one sentence I could actually say in a performance review that describes what I created this month, not what I maintained.