Chaen.ai Work with me

Activity

1Password + Claude Code: Auditing 1,116 Passwords

🗓 ⏱ 5 min read

中文日本語

1Password + Claude Code: Auditing 1,116 Passwords

I moved all 1,116 of my passwords into 1Password and built a Claude Code dashboard to review them one by one. Here is the scoring, the lockout-safe change…

“1Password × CLI” has been going around lately, so I tried it myself. I moved all 1,116 passwords I had into 1Password and built my own dashboard that reviews them one at a time with Claude Code. The score started at 30 and reached 73 after one focused day. This post records what I scored, the order I use to change passwords without locking myself out, and what I checked on the Mac side at the same time.

Why I built a 1Password + Claude Code dashboard

1Password ships its own dashboard. I still built my own, for two reasons. First, I wanted the findings to lead directly into the actual work instead of ending at a list of warnings. Second, I wanted to change the criteria and the priority order to match how I work.

A screen you only look at gets looked at once. A screen where pressing a button starts the work actually gets used. That difference turned out to matter more than any individual check.

1,116 is a large number, but it is what accumulates over years of using the internet. The real problem was never the count — it was having no way to decide where to start. So the first thing I built was aggregation plus a risk score.

One note on the screenshots I posted: the numbers in them are fictional, made for the demo. And I got started quickly because I was able to follow someone else’s post who had done this before me.

The five things I score

The dashboard looks at every single entry through these lenses:

  • Already breached — credentials known to have leaked
  • Reused — the same password used across multiple sites
  • Weak — not strong enough on its own
  • No two-factor — login depends on a single password
  • High-risk AI service accounts — accounts where a lot has accumulated and losing them hurts

The value of a score is that you can see where you stand right now and how much a given fix moves the number. Going through 1,116 entries in registration order guarantees you run out of energy halfway. Sorting by risk so that handling the top few visibly moves the score is what gets you to the end. Going from 30 to 73 happened because the score decided the order of work, not me.

Each card has a “handle these together” button. Pressing it opens the change screens in order of risk. I wanted to separate the time spent deciding what to do from the time spent doing it.

The change order that prevents lockouts

The scariest failure in a password rotation is this: the site now has a new value, and that new value exists nowhere. You are locked out of your own account. So I fixed the order:

  1. Generate the new value and save it to 1Password first
  2. Open the change screen (I do the login and the two-factor step myself)
  3. The tool types the value into the password field
  4. I press the confirm button myself
  5. 1Password is updated to the new value, and the next item opens automatically

The key is that saving comes before changing. If it is already stored, you can recover even if the browser dies mid-flow. The second key is that a human owns the login, the two-factor step, and the final confirm. You can automate those too, and it is faster while it works — but when something goes sideways you have no idea what just happened. I let the machine handle typing and kept the decision.

Password values are never shown on screen and never written into any ledger. Everything runs locally on my own Mac. There are moments where stashing a value somewhere would be convenient, but allowing that hollows out the point of the audit.

The review feature did the most work

The part that helped most was not the score — it was the per-item review. It lines up breached passwords from most to least dangerous, and for each one it gives me:

  • the first thing to do
  • other sites reusing the same password
  • whether to turn on two-factor while I am there
  • whether cancelling the account is faster, if I am not using it

That last one is the sleeper. Once you start an inventory you notice the list is a mix of things worth fixing and things not worth fixing at all. For a service I no longer use, closing the account beats rotating to a strong password. Deciding that one entry at a time is tedious; seeing it laid out makes it instant.

Then one button takes me straight into the work. When investigating, deciding, and doing live on three different screens, you finish at investigating. Putting all three in one place is why this stuck.

Checking the leak entry points on my Mac

While I was at it I audited the leak entry points on the machine itself. Four things:

  • API keys written in plaintext in shell config
  • .env files
  • saved passwords left in Chrome
  • AI agent settings that let it act without confirmation

My honest read: the more heavily you use AI agents, the more exposed this layer is. An agent reads your local environment and acts on it. So a key sitting in plaintext, or a setting that says “run without asking,” becomes live. Password hygiene and agent configuration look like separate topics, but they are the same desk.

In my own case, getting the password side to 73 forced me to fix where the keys were sitting in the first place. Doing one side only means leaking from the other.

What this means for teams in Japan — and for AI companies selling here

In Japan I feel password management still gets treated as a matter of individual diligence. But once people start using AI tools for work, individual account hygiene becomes company risk. AI service accounts accumulate work context, and as more people let agents touch their whole environment, the number of entry points grows with it.

For readers outside Japan, one bit of local context: Japanese organizations tend to move on security when someone can show a current state and a concrete next step in a form that survives internal review. “Be careful, this is risky” does not move anyone. “We are at 30 right now, and clearing the top five gets us here” does. Having the count, the score, and the next action on one screen makes that conversation short.

If you are an AI company trying to grow adoption in Japan, my guess is that showing users a way to see and fix their own state lands better than asserting that you are safe. The reassurance people act on is the kind they can verify on their own machine.

How to apply this

You do not need to rebuild anything. Change the order and you can start today. Before fixing everything, aggregate and score. Fix the rotation procedure around two rules: save the new value first, and let a human press the final button. Keep values out of any ledger. And put AI service accounts — plus local storage spots like shell config and .env files — inside the scope of the inventory. I got this far in a day. What is left is just raising the number.

FAQ

Why build your own dashboard when 1Password already has one?

Two reasons. I wanted the findings to lead straight into doing the work instead of ending at a report, and I wanted to adjust the criteria and priority order to how I actually work. I still use 1Password's own dashboard, but I prefer a screen that executes.

Do the actual password values get handed to the tool or to an AI?

No. The values never appear on screen and are never written into any ledger. Everything runs locally on my own Mac, and I handle the login, the two-factor step, and the final confirm button myself.

How much did the score improve?

It started at 30 and reached 73 after one focused day of work. Sorting by risk and working from the top is what kept me going instead of stalling partway through the list.

Based on a post on XOriginal post ↗

Share this article

XFacebookLINEはてブ

About the author

Masahiro Chaen
Masahiro Chaen Founder & CEO, DigiRise Inc.

CEO of DigiRise Inc. Has supported 600+ companies and 50,000+ people with AI training, consulting and development. GMO AI & Crypto advisor, Google Gemini AI advisor. 200K+ followers on X.