Skip to content
MMW Docs

Connecting MMW gives an agent memory tools. On its own, though, the agent does not know when to use them: it may never look into memory, or it may save everything.

The mmw-memory skill fixes that with one file. The agent loads it by itself when a task fits. It searches memory at the start of a task, reads the conflict and stale statuses correctly, saves decisions together with the reason and never saves secrets.

Skill file: docs.mmwhub.tech/skills/mmw-memory/SKILL.md

For all your projects:

Terminal window
mkdir -p ~/.claude/skills/mmw-memory && curl -fsSL https://docs.mmwhub.tech/skills/mmw-memory/SKILL.md -o ~/.claude/skills/mmw-memory/SKILL.md

For one project only (commit the file and the whole team gets it):

Terminal window
mkdir -p .claude/skills/mmw-memory && curl -fsSL https://docs.mmwhub.tech/skills/mmw-memory/SKILL.md -o .claude/skills/mmw-memory/SKILL.md

Restart Claude Code. Ask it “which skills do you have?” and mmw-memory should be on the list.

Open a new session and give the agent a task in a project MMW already knows something about, for example “add caching to the reports service”. Its first step should be a search on the topic. At the end, if a decision was made, it calls remember with the reason.

If the agent only searches when you ask it to, the skill did not load. Check the file path and restart the client.

The skill follows the best practices:

  • when to search and how to read conflict, stale and superseded;
  • when to save and how to name fact_key so a new version replaces the old one;
  • what never to save: keys, passwords, raw logs, temporary state;
  • when to forget;
  • save the outcome of a task before the final answer.

You can extend the skill. For example, list your team’s workspaces at the end of your copy:

## Workspaces
- decisions: architecture and process decisions
- docs: facts from documents, always with source_id