A Claude skill that catches the mistake before the client does — built for operators who cannot afford to find out later
The email was already sent. That is the part that never gets easier. You spent forty minutes building the deliverable, another ten cleaning it up, and then you hit send because the client was waiting and you were already behind. Three hours later they write back. The attachment was missing a section. Or the numbers did not match the brief. Or the tone was wrong for the audience you both discussed on the call last Tuesday. You knew that call. You just forgot it at the moment it mattered.
This is not a focus problem. It is a systems problem. Business operators working with AI are producing more output than they ever have before. More drafts. More proposals. More emails. More reports. The speed is real. So is the gap. When you move fast, the check step gets skipped. Not because you do not care. Because there is no check step built into the process. The approval moment sits between the work and the send button, and right now it is just you, tired, clicking through.
This book gives you one tool. A Claude skill called the Agent Clearance Skill. You paste it once. From that point on, before anything ships, before anything gets sent, before anything lands in front of a client or a boss or a partner, Claude runs it through a structured review and tells you exactly what is wrong. Not in theory. Right now, on the thing you just built. By the time you finish this book you will have that skill installed in Claude Projects and running on real work. That is not a concept. That is a working checkpoint that did not exist in your process yesterday.
She had been working with AI for six months when it happened. Her name does not matter. What matters is the output. She had asked Claude to write a client-facing proposal for a mid-size logistics company. Claude produced it. It was good. She read the first paragraph, scrolled to the pricing section, saw the number she had given it, and sent the PDF.
The client called the next morning. The proposal used the wrong company name throughout. Not her company. The previous client. Claude had carried it over from a session she had not closed properly. Twenty-two instances of the wrong name in a twelve-page document. Professional. Polished. Wrong.
This is the failure pattern that nobody talks about when they talk about using AI in business. They talk about speed. They talk about output quality. They do not talk about what happens in the handoff. The space between Claude finishing the work and the work reaching the person who will judge you on it. That space is where the damage happens. Not because the AI made a catastrophic error. Because nobody checked.
The problem is not the mistake. Mistakes exist in every process. The problem is that operators who use AI have removed the natural pauses that used to create review moments. When you wrote everything by hand, the act of writing was the review. You read while you wrote. You caught things because the process was slow enough to surface them. AI broke that. It produces the thing before you have processed what you asked for. You are always one step behind the output.
The Agent Clearance Skill is built for this exact gap. It does not replace judgment. It restores the checkpoint that speed removed. Before anything ships, you run the output through this skill. Claude reviews it against four criteria. It tells you what passes. It tells you what does not. It names the specific line or section that needs attention. Then you fix it. Then you send it.
The skill runs in Claude Projects. That is Anthropic's feature that lets you give Claude a persistent set of instructions that live in a project space and apply every time you open that project. You do not re-explain yourself. You do not re-prompt. You open the project, paste the output, ask Claude to clear it, and Claude already knows what to look for.
Here is the exact prompt. Open Claude. Go to Projects. Create a new project and name it something you will recognize. Paste this into the project instructions field.
1. RECIPIENT MATCH — Does the content address the correct recipient by name, role, and context? Flag any name, company reference, or role that does not match what I have described about this recipient.
2. BRIEF ALIGNMENT — Does the content fulfill what I asked for? Compare it against any brief or instructions I have provided. If I have not provided a brief, ask me for one before reviewing.
3. TONE CHECK — Is the tone appropriate for the relationship? I will tell you the relationship type (new client, existing client, internal, vendor, public). Flag any line that feels mismatched.
4. COMPLETENESS — Is anything missing that was required? Flag sections that were promised or implied but not delivered.
After reviewing, give me three outputs: - CLEAR: list each criterion and whether it passed - FLAG: name every specific line or section that failed, and why - FIX: for each flagged item, write the corrected version
Do not summarize. Do not editorialize. State what the record shows and give me the fix.
Before the next chapter, do this one thing: open Claude Projects right now, create a project called Agent Clearance, and paste that prompt into the instructions field.
Here's What We Built. We run a publishing operation that produces ebooks on a schedule. Every book that goes out passes through a clearance step before it moves to the next stage. The pipeline we built does not allow a draft to advance without a review. The skill you just installed is the same gate we use.
The consultant had been on the call. He heard the client say it clearly. "Make sure the language is accessible. Our board is not technical." He nodded. He took notes. He opened Claude and typed a long prompt that described the deliverable in detail. He did not type the part about the board. He forgot it the way you forget the one thing that mattered.
Claude produced a document. Precise. Data-heavy. Full of terms that would have made the board feel like outsiders in their own meeting. The consultant reviewed it for accuracy. Everything checked out. He sent it.
The feedback came in the form of a silence. The kind where the client stops scheduling calls.
What broke here was not the AI. Claude gave him exactly what he asked for. The brief he typed was accurate. It was just incomplete. The gap between what he remembered from the call and what he put into the prompt was small. The consequence was not.
The second criterion in the Agent Clearance Skill addresses this directly. Brief Alignment. But Brief Alignment only works if the brief exists somewhere Claude can see it. This chapter is about building that brief before you build the deliverable. Not after. Not as an afterthought. Before.
The move is simple. Before you ask Claude to build anything you intend to send to another person, you open the Agent Clearance project and you log the brief. One paragraph. You write what the deliverable is, who it is for, what they told you about their needs, and what a pass looks like. You paste that as a message at the top of the session. You label it BRIEF. Then you go build the deliverable, in the same session or another, and when you bring it back for clearance, Claude has the brief in context.
Here is the exact prompt addition for logging a brief. You do not change your project instructions. You use this as the opening message of every Agent Clearance session.
Deliverable: [what you are building — one sentence] Recipient: [name, company, role, and any relationship context you have] Their stated needs: [what they told you, in their words if you remember them] Tone: [new client / existing client / internal / vendor / public — pick one] What a pass looks like: [how you will know this deliverable did its job]
This brief is active for this session. When I ask you to clear something, compare it against this brief.
The thing that most operators skip is the last line. What a pass looks like. They describe the deliverable. They describe the recipient. They stop there. But a clearance agent cannot tell you if something works if nobody defined what working means. Write that line. Be specific. "A pass looks like the board reading this without needing to ask a single clarifying question." That is specific. That is something Claude can evaluate against.
Before the next chapter, do this one thing: open your Agent Clearance project, start a new session, and fill out the BRIEF LOG for the next deliverable you are actually building this week.
Here's What We Built. Our publishing pipeline runs this exact sequence every time a book is produced. Brief first. Build second. Clear third. Ship fourth. The sequence is not a concept. It is a live operation. You just installed step one.
The agency owner had a template. She had spent two weeks building it. A client onboarding email sequence, five messages, designed to move a new client from signed contract to first meeting without her having to write a single word manually. Claude would generate each message from a short prompt. She would review and send. She was proud of it. It was good work.
Six months later she had a problem she could not name. Clients were completing onboarding. They were showing up to the first meeting. And something was off. The meetings felt like first meetings with strangers instead of first meetings with people who had been communicating for a week. The warmth was correct. The information was accurate. The tone was wrong in a way she could not measure.
She pulled the sequence and read it the way a client would read it. Start to finish. All five messages. By message three she saw it. The sequence was written for someone the agency owner had never met. A generic client. Professional. Responsive. Organized. The clients who hired her were none of those things by the time they signed. They were nervous. They were investing money they were not entirely sure about. They needed a different kind of reassurance than the one the template provided.
The tone check criterion in the clearance skill exists for this. But tone is the hardest criterion to define, because tone is relational. It changes based on where the relationship is, not just who the relationship is with. A message to an existing client of three years sounds different from a message to a client who signed forty-eight hours ago. Same company. Different moment. Different tone.
Here is how you feed that context into the skill. When you log your brief, the tone field does not just take a category. It takes a sentence. Not "new client." Instead: "New client, forty-eight hours post-signature, nervous about the investment, needs reassurance more than information." That sentence changes what Claude looks for during the tone check.
Here is the exact prompt addition for a tone sentence. Replace the single-word tone field in the BRIEF LOG with this format.
Example: New client — signed forty-eight hours ago, nervous about cost, needs to feel like they made the right choice more than they need to know the next steps.
One more thing about the tone check. It will flag lines that are technically correct and emotionally wrong. A line like "Please review the attached onboarding document and complete all required fields before our call" is accurate. It is also cold to a nervous client who just spent money. Claude will flag it. That flag is the skill working.
Before the next chapter, do this one thing: take the last email you sent to a client, paste it into your Agent Clearance project, run a tone check with a brief that includes a full tone sentence, and read what Claude flags.
Here's What We Built. Every book in this series passed through a tone check at every draft stage. The voice you are reading was not an accident. We ran the same check on this sentence before it shipped.
The freelance strategist submitted the proposal. It was ten pages. It covered the engagement model, the timeline, the deliverables, and the pricing. The client signed off on the pricing within an hour. Then they asked a question the strategist had not anticipated. "What happens if this does not work?"
There was no answer in the proposal. Not because the strategist had not thought about it. She had thought about it. She had a whole framework for measuring success and adjusting the engagement. She just never wrote it down in the document because she assumed they would cover it in the kickoff call.
They did not get to the kickoff call. The client went quiet after asking the question. The silence lasted eleven days. Then the contract was cancelled.
The completeness criterion in the clearance skill is designed for this. It looks for sections that were implied, promised, or expected but not delivered. The problem is that Claude cannot know what the client expected unless you tell it. This chapter is about giving Claude the information it needs to catch the missing piece.
The move is adding one more field to your brief log. Expected sections. You list, specifically, every section or element this deliverable is supposed to contain. Not every section you put in. Every section the recipient expects to find. Those are sometimes the same. They are not always the same.
Here is the exact prompt addition for expected sections. Add this field to your BRIEF LOG.
Example: Executive summary, scope of work, timeline, pricing, success metrics, risk mitigation, next steps.
The completeness check is the quietest criterion in the skill. Recipient match catches visible errors. Tone check catches emotional mismatches. Brief alignment catches strategic drift. Completeness catches the thing that was never there to catch. The absence. The gap. The section that everyone assumed was handled.
Run the expected sections list every time. Even for short deliverables. Especially for short deliverables. A three-paragraph email can be missing one sentence that the recipient needed. That sentence is not going to surface in a tone check or a brief alignment check. Completeness is the only criterion looking for it.
Before the next chapter, do this one thing: take a proposal or report you sent in the last thirty days, write the expected sections list for it from memory, paste both into your Agent Clearance project, and ask Claude to run a completeness check.
Here's What We Built. Every product we have released was audited for completeness before it shipped. The ones that kept getting positive reviews months after launch were the ones where we ran this check before we hit send. The ones we skipped it on — we knew within a week.
The marketing manager had everything set up. The campaign email was ready. The subject line was tested. The list was segmented correctly. She had used the clearance skill for the first time. Claude had flagged two things. She had fixed both. She read the email one more time.
Then she made a small change.
She adjusted the opening sentence because she thought it sounded better. Tightened the closing. Added a stat she had seen that morning. She did not run clearance again. The change was small. The whole thing took twelve seconds.
The stat was wrong. Not drastically. Off by a year. But the source it linked to was outdated, and three people in the segment noticed, and two of them replied to tell her.
The last twelve seconds. That is where the clearance skill either works or fails. Not in the prompt. Not in the setup. In the discipline of running clearance after the last change, not before it.
This chapter is about making that discipline automatic. The way you do it is by building a send rule and attaching it to the project. Not a rule you try to remember. A rule you write down and paste as the final instruction in your Agent Clearance project.
Here is the send rule. Add it to the bottom of your project instructions.
The send rule is not about distrust. It is about designing a process that works when you are tired, rushed, and already thinking about the next thing. Which is when you are always sending.
Before you close this book, do this one thing: add the SEND RULE to your Agent Clearance project instructions right now, before you send anything else today.
Here's What We Built. We built this exact gate into our own production process. Nothing ships until it clears the final check after the final edit. The book you are reading cleared it. That is the only reason it is in your hands.
You finished the book. That means you have the Agent Clearance Skill installed, a brief log structure that catches misalignment before it ships, a tone check that runs on relationship context not just category, a completeness audit that surfaces the missing section, and a send rule that enforces the discipline when discipline is hardest.
Not concepts. A working checkpoint system running in Claude Projects right now.
The clearance process in this book is the same process we run on every piece of content we produce. The gate is real. The pipeline is live. The book you are reading passed through it. That is not a claim. That is what the record shows.
Everything we have built from this process is at aixsweb.com. Come find what we built. Then build something of your own.
aixsweb.com
---
Practical guides for small business owners building online the right way.
No spam. Unsubscribe anytime.
You're in. Next guide coming your way.