Careful Mode by SOLIDS FOR YOU

Example session with this skill installed

careful mode. Update the prices in this product sheet to the new catalog and draft the customer email announcing them.

Read your context and instructions
Compiled the careful mode by

(the updated sheet and the email draft, then:)
Careful mode: reviewed 3×

Correctness: checked all 24 prices against the catalog; fixed 1 (Linen Shirt is ₹1,199, not ₹1,299).
Completeness: sheet and email both done; email under 150 words as asked; nothing extra added.
Adversarial: email and sheet now use the same product names; sending the email can't be undone, so it is a draft for you to send.
Not verified: the live store page (no access).

Connects securely to your tools. The creator never sees your data.

About this skill

Careful mode makes Claude check its work three times before handing anything over: a reply, a file, a commit, a message, a price change. Turn it on with "careful mode" and off with "careful mode off".

Each review has its own job, so they don't just repeat each other:

Correctness: is it right? Every price, name, date, link and number is checked against the source again, not from memory. Code gets run and traced with edge cases. Calculations get redone a second way.
Completeness: is it everything? Each requirement from your request gets ticked off. Nothing missing, nothing you didn't ask for, no leftover TODOs or placeholders.
Adversarial: how could it go wrong? Claude reads the work like a strict critic, checks that names and prices match everywhere, and asks before anything that can't be undone.

If a review finds a problem, Claude fixes it and starts again at review 1, because a fix can bring a new mistake. Every handover ends with a short record of what was checked, what was fixed, and what couldn't be verified, so you can see the reviews did real work.

Use it for anything where a mistake costs money or trust: prices, invoices, customer emails, contracts, production deploys.

There's a paste-in version (careful-mode.md) for any Claude chat. Free and open source (MIT), by SOLIDS FOR YOU.

It makes Claude slower and uses more tokens, because every piece of work is reviewed three times. It can only verify what it can see or run: if it has no access to a source (a live site, a private system), it says so instead of guessing. It doesn't replace a professional review for legal, tax or medical work.

Open docs

How to use

1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.

Reviews

3.0

1 review

This skill has been used 10 times.

Frequently Asked Questions

Say "careful mode" (or "triple check", "be extremely careful"). It stays on for the rest of the session until you say "careful mode off".

No. In the third review it checks the side effects of every action, and anything that can't be undone waits for your confirmation.

The skill is free. Three reviews use more tokens and time than one, so use it for work where a mistake is expensive.

The work can be handed to helpers, but the three reviews are always done by the main model, and anything a helper returns is reviewed like Claude's own work.

Careful Mode by SOLIDS FOR YOU