Description
how-it-works is a skill from the claude-mem project that explains how claude-mem captures observations, when memory is injected, and where the data lives. claude-mem gives Claude Code a persistent memory across sessions. Every Read, Edit and Bash action that Claude takes becomes a compressed observation. Observations are summarized when a session ends, and the relevant ones are injected automatically into later prompts, so the next session starts with context from the previous one without re-explaining the codebase or re-discovering past decisions. Memory injection begins in a project's second session. The first session in a new project seeds the memory, and an optional learn-codebase command can load the whole repository into memory in one pass, which takes about five minutes. All data stays in a claude-mem folder in the user's home directory on the local machine, including the SQLite database, the vector index, logs and settings. Data leaves the machine only in calls to the AI provider the user has configured for compression, which can be Claude, OpenRouter or Gemini. A local install contacts cmem.ai once at sign-up to create the sign-in link, and nothing else is sent there.…
Highlights
- Captures observations from Read, Edit and Bash
- Summarizes observations at session end
- Injects relevant memory into later sessions
- Data stored locally, cleanly removable
Ratings & Reviews
No ratings yet
Be the first to rate how-it-works.
People also visit
continuous-learning
[DEPRECATED - use continuous-learning-v2] Legacy v1 stop-hook skill extractor
unified-memory
Share durable, inspectable context and handoffs between Claude, Codex, Hermes
remember
Explicitly save important knowledge to auto-memory with timestamp and context
how-it-works FAQ
What does the how-it-works skill explain?
How claude-mem records observations from Claude Code sessions, when that memory is injected, and where the data is kept.
How do I install it?
Run: npx skills add https://github.com/thedotmack/claude-mem --skill how-it-works
When does memory start to appear?
From a project's second session. The first session seeds the memory, and an optional learn-codebase command can load the whole repository.
Where is the data stored?
Everything stays in a claude-mem folder in the home directory on the local machine, including the database, vector index, logs and settings.
What leaves my machine?
Calls to the AI provider configured for compression, such as Claude, OpenRouter or Gemini. A local install contacts cmem.ai once, at sign-up.
What are the best how-it-works alternatives?
Popular alternatives to how-it-works include ck, context-budget and continuous-learning. See all how-it-works alternatives