Adam Daniel

A map of what Claude actually remembers

2026-06-25T13:00:00+00:00

Claude’s memory is scattered across more places than most people realize — and which memories follow you around depends entirely on how you’re using it. A standalone chat draws on your account-level memory; a chat inside a Project draws on that Project’s own store instead. Claude Code running in WSL keeps a set of notes your Windows-side Claude Code never sees. Claude Code on the web reads only the repo’s CLAUDE.md and nothing else. Cowork has its own per-project memory again.

I kept having to re-derive all of this from memory — ironically — every time someone asked me “wait, will it remember that?” So I built a small thing to answer the question once: an interactive map of every place Claude stores something, and how it flows in and out.

Check the ways you actually use Claude and the diagram composes itself — no pre-baked variants, it’s generated live from your selection. There’s a scope lens to ask the two questions that usually matter (“what follows me across projects?” vs. “what stays put?”), and a toggle for the full, official Anthropic terminology when you want the precise names rather than the friendly ones. It’s one self-contained page; nothing you check is sent anywhere.

The permanent home is here — Claude Memory Map — and it’s embedded below so you can poke at it without leaving the page:

Trouble with the embed? Open the full-page version →

The whole thing is open source — the code is on GitHub — and every store, edge, and label traces back to documented Anthropic behavior (where the docs and the shipping app disagreed, the app won). If you spot something that’s drifted out of date, a pull request or a note is very welcome.

Quoting Anthropic: Opus 4.8 Safety “somewhat less robust”

2026-05-28T19:35:00+00:00

Agentic safety. Although it shows improvements in some areas (such as refusing malicious requests), we found Opus 4.8 to be somewhat less robust than Opus 4.7 in several agentic contexts (such as vulnerability to prompt injection attacks). However, the application of our safeguards closes the gap between the models in practice. […]

- Anthropic, System Card: Claude Opus 4.8

Introducing GHA-bench

2026-05-13T12:51:00+00:00

GHA-bench is a benchmark and a set of evals for how well different coding agents author and test GitHub Actions.

How it works

Agents (currently a variety of Anthropic models set to various effort levels, driven by Claude Code) are given set of tasks they must automate using GitHub Actions, either using a particular scripting language or whichever they want.* They must use Test-Driven Development (TDD)– basically “write tests first, and don’t come back until they all pass”.**

A panel of judges (Google Gemini and Claude Haiku) then evaluates the comprehensiveness of the tests and the quality of the code.

Which model, effort level and scripting language should you use?

Adjust the sliders according to your priorities.

Duration 17.5%

Cost 17.5%

Tests Quality 40.0%

Code Maintainability 25.0%

Model	Language	Duration	Cost	Tests	Code

* When allowed to choose, the agents always choose Python.

** Agents run their tests locally in a container that leverages nektos act to emulate a GitHub-hosted runner.