---
title: "claude: recap"
url: https://perrotta.dev/2026/04/claude-recap/
last_updated: 2026-07-29
---


Now that frontier models from Anthropic have context windows of up to 1M tokens,
I constantly find myself forgetting what each session is about whenever I have
4+ terminal tabs open.

I know, I could simply [name sessions]({{< ref
"2026-04-12-claude-code-naming-sessions" >}}), but I often don't.

Can we throw AI at the issue? Of course, Gen AI solves problems generated by AI
(pun intended).

There's this built-in ~~command~~ skill: `/recap`:

> Generate a one-line session recap now

200K+ tokens later, `/recap` is my hope to remind me of what I was doing the day
before.