Will it change my code without asking?

The review never does. The workflow does, on its own branch, and only after you confirm the phases it shows you. You merge the pull request or you do not.

What happens if I close my laptop halfway through?

The run stops with the session. Reopen that session with claude --resume and ask Claude to relaunch the workflow, and finished agents return their saved results. Start a fresh session instead, and it runs from the top.

What if it finds ninety gaps?

Then the report has ninety entries and the workflow closes the Blockers and the 30-day ones. Later gaps stay listed. Re-run the review after the merge and they are what is left.

Video guide

Find every gap in your product tonight. Close them by morning.

The run in the video launched 190 agents and worked for 16 hours. The part worth copying is the audit inside it: a skill that reads your repo the way nine department heads of a real company would, and writes a gap report an agent can act on. This page gives you that skill, the prompt that runs it, and the workflow prompt that closes what it finds.

Watch the videoAbout a minute and a half, on Instagram.

Departments in the audit: 9
Prompts to paste: 2
Our run, unattended: 16 h
The skill download: Free

What you will have when you finish

A readiness report with a verdict, and a pull request that closes the gaps it found.

A report at docs/commercial-readiness.md: a verdict, a scorecard, and every gap with the file to open, the steps to take, and how to verify it
Nine lanes checked at once: legal, risk and compliance, security, operations, finance and billing, marketing, customer relations, sales, and people
A branch where a workflow closed every Blocker and 30-day gap, one agent per gap, with a check between the tiers
A second run of the review that moves the closed gaps into the Closed table
One pull request to read over coffee, and nothing deployed

Four steps

Steps one and two take five minutes. Three and four are Claude Code's time, not yours.

Claude Code on a paid plan, in a repo you can afford to let agents change. On Pro, turn on Dynamic workflows in /config before step four.

01
Open a terminal in the project you want reviewed and start Claude Code.
cd your-project claude
Use the repo that holds the product you charge for, or want to. The review reads code, config, docs and the deployed pages, so it needs the repo, not a description of it.
02
Download the commercial readiness skill and unzip it into your skills folder. Claude Code picks it up the next time you type a slash.
Download the skill~/.claude/skills/commercial-readiness/SKILL.md
Sixteen files, no account, nothing installed beyond the folder. On Windows the folder is C:\Users\you\.claude\skills. Type a slash in Claude Code and look for commercial-readiness in the list to confirm.

Paste the review prompt. Claude Code reads the skill, writes a brief about your product, sends nine agents out at once, and comes back with a verdict and a report.

Prompt one, the review

Use the commercial-readiness skill to review this repo. Evidence scope is repo plus live: read the code, config and docs, fetch the deployed marketing and legal pages, and check DNS. Never log in to anything and never change anything. Write the report to docs/commercial-readiness.md, copy the brief and the nine lane files into docs/readiness/, and build the report page. Then tell me the verdict, the blocker count, and the three gaps to close first.

Read the verdict before you go on. Any Blocker means not ready to charge. The three gaps it names first are the ones the workflow hits first.

Paste the workflow prompt, confirm the phases Claude Code shows you, and leave it running.

/workflows

Prompt two, the workflow

Use a workflow to close the gaps in docs/commercial-readiness.md. Create a branch named readiness/close-gaps from the current branch and do all of the work there. Phases and gates:

1. Blockers. One agent per gap whose metadata line says Severity: Blocker. Each agent reads only its own gap and closing brief, makes the change, runs that gap's Verify line, and commits with the gap ID first in the subject. Gate: every Blocker's Verify line passes and the project's test command is green.

2. 30-day. The same, one agent per gap whose metadata line says Severity: 30-day. Gate: the same.

3. Review. One agent re-reads every closed gap against its closing brief and reverts any change that does not match the brief. Gate: the reviewer lists each gap as closed or reverted.

4. Re-run. Run the commercial-readiness skill again and write the new report over docs/commercial-readiness.md, so closed gaps move to the Closed table. Gate: zero Blockers remain.

5. Pull request. Open a pull request against the branch I started from, with the Closed table and the reverted list in the description.

Rules for every agent: no deploys, no database migrations, no billing or payment commands, and no new paid service without naming its monthly cost in the pull request. Gaps marked Later stay open. Keep every gap ID unchanged.

Stay in the same session; a fresh session starts the run over. A run pauses at any permission prompt, so allow the tools your agents need before you start, or run in a mode that does not ask. In the morning, read the pull request, not the transcript.

Every agent counts against your plan's usage. A review is nine agents. Closing twenty gaps is twenty more plus the reviewer and a second review. Run prompt one on its own first and see what it costs you.

The specifics

What the skill checks: Legal pages, data handling, security, monitoring, billing, marketing, support, checkout, and whether the company survives you taking a week off.
How it scores: Against your stage, not a public company: pre-launch, first paying customers, or about $10k MRR. Gaps come out as Blocker, 30-day or Later.
What our run did: 190 agents, 16 at once at the peak, 5,035 tool calls and 23 million tokens, from 9:31 AM to 1:32 AM the next morning.
What it produced: 148 commits, 44,198 lines across 338 files, 675 tests, an 87 finding readiness report and one pull request. Nothing deployed.
What it touches: The review changes nothing. The workflow works on a branch named readiness/close-gaps and never deploys, migrates a database, or bills anyone.
What it costs: The skill is free. The agents count against your Claude plan, and a full closing run uses a lot of it.

Questions people ask

Will it change my code without asking?
The review never does. The workflow does, on its own branch, and only after you confirm the phases it shows you. You merge the pull request or you do not.
What happens if I close my laptop halfway through?
The run stops with the session. Reopen that session with claude --resume and ask Claude to relaunch the workflow, and finished agents return their saved results. Start a fresh session instead, and it runs from the top.
What if it finds ninety gaps?
Then the report has ninety entries and the workflow closes the Blockers and the 30-day ones. Later gaps stay listed. Re-run the review after the merge and they are what is left.

Want a hand with your first run?

GraniteAI builds and runs AI systems for small and midsized businesses, out of Manchester, New Hampshire. We wrote the skill and ran the review on our own product first. If you would rather do your first one with someone who has read a few of these reports, book a session below.

Book a session See every tool