Save Progress — AI skill for Claude Code

**A Claude skill that stops your AI forgetting every correction you give it.**

You correct Claude. It thanks you. Tomorrow it does the same thing again.

The moment your AI most needs to learn from you, the moment you tell it it got something wrong, is the one moment nothing captures. So you explain the same preference on Monday, fix the same mistake on Thursday, and start from zero next week.

This skill runs at the end of a session. You type two words. It reads back over the conversation, works out what is worth keeping, and files each thing where it belongs, without overwriting anything already there.

save progress

That is the entire interface.

What makes it different

Most memory systems sort what they learn by **topic**. This one sorts by **how long it should last**.

That single change is what stops a memory file turning into landfill.

Bucket	What it is	How long it lasts
A durable rule	A preference that should apply to every future job	Forever, until you change it
A one-off override	A choice true for this one piece only	This piece only. Never becomes a rule
Project state	What shipped, what is open, what you are waiting on	Until the project ships
A lesson	A mistake that should not repeat	Permanent, as a warning

The safeguard

Before anything is filed as a permanent rule, one question:

**Would I want this applied to every future job?**

If no, it is an override. It goes in project notes and never becomes a rule.

Promoting a one-time choice into a global rule is the failure that quietly poisons a memory system. It is invisible for weeks, then suddenly every output carries a preference you expressed once, about one thing, months ago.

It will not clobber your work

Every write follows the same protocol. Read the file first. Find any existing entry on the same topic. Update that entry in place rather than adding a duplica