Remember
Write a memory to the mematrix timeline. Use when user invokes /mematrix:remember, /mematrix:write, or asks to "remember this", "save this for later", "make a note", "don't forget", "memorize this", or wants to persist information from the current conversation.From its SKILL.md
npx -y skills add themarchrain/mematrix --skill rememberAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 27 days oldThe repository was created 27 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.3 KB, 711 tokens by cl100k_base, as published. Nobody here has run it
Mematrix Remember — Write memories to timeline
Write a memory to ~/.mematrix/timeline/YYYY-MM-DD.md with dedup checking.
Memory architecture
~/.mematrix/
global/ ← Stable long-term facts (profile, technical, preferences)
timeline/ ← Daily episodic memories (YYYY-MM-DD.md)
.mematrix/ ← Project-level (if workspace was used)
Two modes
Mode A: Explicit content
User provides the memory text directly:
/mematrix:remember I use pnpm instead of npm for all Node.js projects
Use the text as-is.
Mode B: Session scan
User provides no text:
/mematrix:remember
Review the current conversation thoroughly and extract what's worth keeping:
- Decisions made and their rationale
- Technical details, bugs encountered, solutions found
- Project context, architecture notes
- User preferences or constraints expressed
- Tools, commands, or workflows discovered
- Any info the user's future self (or another agent) would need
Format EACH extracted insight as a separate memory entry.
Write procedure (both modes)
1. Dedup check (do this BEFORE writing anything)
For EACH candidate memory:
- Read ALL files in
~/.mematrix/global/ - Read the last 30 days of
~/.mematrix/timeline/ - Compare the candidate against existing content semantically, not textually:
| Situation | Action |
|---|---|
| High similarity — same fact already recorded | SKIP. Tell user "Already in [file]" |
| Related but adds detail — extends existing info | MERGE. Append to the existing entry |
| Genuinely new — no match found | WRITE. Add as new entry |
| Uncertain — moderate similarity, ambiguous | WRITE. Conservative: a duplicate costs tokens; a missing memory costs context |
Examples:
- "I use pnpm" vs "Package manager: pnpm" → SAME, skip
- "I prefer dark themes" vs "I use VS Code with Dark+ theme" → RELATED, merge
- "Debugged API timeout" vs "Fixed connection pool config" → DIFFERENT, both keep
2. Write to today's timeline file
File: ~/.mematrix/timeline/YYYY-MM-DD.md
Create the file with an # YYYY-MM-DD header if it doesn't exist.
Entry format:
## HH:MM — Short descriptive title
What happened, what was decided, why, context, constraints.
**Tags:** #tag1 #tag2
Use the actual current time for the timestamp.
3. Global update check
After writing to timeline, evaluate: does any extracted/changed info update a global file?
- New language/framework learned → update
technical.md - Changed preference → update
preferences.md - New role or life change → update
profile.md
If yes, update the relevant global file AND tell the user.
Principles
- Conservative dedup: when similarity is ambiguous, WRITE. Missing context is worse than a near-duplicate.
- Global is precious: global files are stable identity. Update them only when something genuinely changed.
- No auto-memory: only write when the user explicitly invokes this command.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.