Update the right memory.md at the end of a significant exchange. Use whenever a durable fact appeared - a decision, a preference or rule from the user, a change in the state of a project or topic, a reusable fact (path, URL, command, convention). Skip it for one-off questions and chit-chat.
Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Memory?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/makushingu-memory)
---
name: memory
description: Update the right memory.md at the end of a significant exchange. Use whenever a durable fact appeared - a decision, a preference or rule from the user, a change in the state of a project or topic, a reusable fact (path, URL, command, convention). Skip it for one-off questions and chit-chat.
---
# Memory
The workspace has three memory levels, four with agents. Each `CLAUDE.md` imports its `memory.md` (`@memory.md`), so
whatever you write here is loaded automatically in the next session of that topic.
## When
At the end of an exchange that produced something worth knowing next time:
- a decision or a chosen direction;
- a preference or rule stated by the user ("always…", "never…", "from now on…");
- progress that changes the state of the work (done, blocked, next step);
- a reusable fact: path, URL, command, account name, convention, gotcha.
Nothing durable (a quick question, reading a file, a failed attempt with no lesson) -> write nothing.
## Where (pick the narrowest level that is true)
0. **Agent** - `<workspace>/projects/<project>/topics/<topic>/agents/<agent>/memory.md`, only when the
session runs in an agent folder (the working directory ends in `agents/<agent>`): what this agent
learned doing its role (sources it uses, what worked, its own conventions). A fact true for the whole
topic goes one level up, in the topic memory.
1. **Topic** - `<workspace>/projects/<project>/topics/<topic>/memory.md`: the current mission. Most
common; the session's working directory tells you the topic.
2. **Project** - `<workspace>/projects/<project>/memory.md`: true for every topic of the project.
3. **Global** - `<workspace>/memory.md`: rules and preferences valid everywhere. Rare.
The workspace root is printed by `maxxing status` (here `{{WORKSPACE}}`).
## How
- Short factual bullets in the matching section; add a date when it helps: `- 2026-03-02: …`.
- **Replace** a fact that became wrong; never keep two contradictory versions.
- Merge with an existing bullet rather than adding a near-duplicate.
- No conversation logs, no step-by-step narratives: only what is reusable.
- **Never write secrets** (tokens, passwords, API keys). Store them in Settings -> Keys and write
only the key *name* in memory ("GitHub token: key `GITHUB_TOKEN`").
- Instructions that should always apply (how to work on this project) belong to `CLAUDE.md`, not
`memory.md`. Edit `CLAUDE.md` only when the user asks for a lasting instruction; keep its first
paragraph (the description shown in the UI) intact.
- Keep the edit small; tell the user in one line what you recorded ("Noted in topic memory: …").
