Skip to content

REMEMBER-md-guide.md

Latest commit

 

History

History
330 lines (218 loc) · 10.1 KB

REMEMBER-md-guide.md

File metadata and controls

330 lines (218 loc) · 10.1 KB

REMEMBER.md Guide

Your custom instructions for how your Second Brain captures, processes, and organizes knowledge.

What is REMEMBER.md?

REMEMBER.md is a user-editable file at your brain root (e.g., ~/remember/REMEMBER.md) where you define custom rules for how Remember works. It uses plain markdown — no schemas, no config syntax, just natural language that Claude reads and follows.

Key principles:

  • You own it — Remember never auto-modifies this file (unlike Persona.md)
  • All sections are optional — leave a section empty or remove it entirely to use defaults
  • Additive by default — your instructions augment built-in behavior, not replace it
  • Zero config works — if you never touch REMEMBER.md, Remember works exactly as before

REMEMBER.md vs Persona.md

REMEMBER.md Persona.md
Who writes it You (the user) The system (/remember:process)
Purpose What you WANT the brain to do What the AI OBSERVES about you
Nature Prescriptive (instructions) Descriptive (patterns)
Auto-modified Never Yes, during processing

Precedence: REMEMBER.md > Built-in Defaults > Persona.md

Sections Reference

## Capture Rules

Define what to always or never capture, and set thresholds.

Example:

## Capture Rules

### Always Capture
- Client meetings and who attended
- Decisions with rationale (especially "why NOT" something)
- Book/article recommendations from anyone
- Health observations (symptoms, energy levels, sleep quality)

### Never Capture
- Casual greetings or small talk
- Debugging sessions (unless a reusable lesson emerges)
- Code snippets without context (code belongs in repos, not the brain)
- Venting — if I'm just frustrated, don't save it unless I explicitly say so

### Capture Thresholds
- Ignore messages under 5 words unless they contain a name or decision
- For long sessions, prioritize: decisions > tasks > learnings > context

How it works: During brain dumps and /remember:process, Claude checks these rules FIRST before applying default routing. "Never Capture" items are skipped even if they'd normally be captured. "Always Capture" items are saved even if they seem too minor by default.

## Processing

Control how content is routed, formatted, and tagged.

Example:

## Processing

### Routing Overrides
- Anything about "Impact3" → Projects/impact3/ (always, even if it sounds like an Area)
- Health data → Areas/health.md, but mental health → Areas/personal-growth.md
- "Reading list" items → Resources/books/ even if no URL is provided

### Output Style
- Journal entries: bullet points grouped by project, not prose paragraphs
- People interactions: include direct quotes when available
- Decisions: always include "Alternatives considered" and "Why this choice"
- Tasks: include estimated effort when I mention it (⏱️ tag)

### Tagging
- Auto-tag client meetings with `billable` and the client name
- Tag all decisions with `decision` + domain (tech, business, personal)
- Tag learnings from mistakes with `lesson-learned`

How it works: Routing Overrides apply before the default routing table. Output Style affects how content is formatted when written. Tagging rules are applied to frontmatter tags.

## Custom Types

Define entity types beyond the standard PARA structure.

Example:

## Custom Types

### Meeting Notes (→ Projects/{project}/meetings/)
When processing a meeting or call:
- Create: `Projects/{project}/meetings/YYYY-MM-DD-topic.md`
- Sections: Attendees, Key Points, Decisions, Action Items
- Always link attendees to People/

### Habits (→ Areas/health.md, section: ## Habit Tracker)
When I mention starting, stopping, or tracking a habit:
- Format: `| Date | Habit | Status | Notes |`
- Track: exercise, meditation, reading, sleep schedule

### Bookmarks (→ Resources/bookmarks.md)
When I share a URL without much context:
- Don't create a full Resource file
- Instead, append to Resources/bookmarks.md as: `- [Title](url) — one-line description ({date})`

How it works: During classification, Claude checks if content matches a Custom Type before falling through to the default routing table. Each Custom Type specifies where and how to store the content.

## Connections

Define auto-linking rules and people context.

Example:

## Connections

### Auto-Link Rules
- Any mention of "Impact3" + a person → link both to Projects/impact3/
- "Family" topics → always cross-link to People/ entries for family members
- Tech decisions → link to both the Project and Notes/decision-*

### People Context
- Maria = my wife (don't re-explain this in every People/ interaction)
- Alex = cofounder at Impact3 (link to Projects/impact3/ automatically)

How it works: When creating or updating notes, Claude uses these rules to add wikilinks automatically. People Context prevents redundant "who is this person" explanations in interaction logs.

## Language

Control multilingual capture and processing behavior.

Example:

## Language

- I switch between English and Romanian freely
- Capture in whatever language I used
- Process/organize using English for headers and tags
- Romanian is fine for content/body text

How it works: Injected into both brain dump hooks and /remember:process context. Tells Claude how to handle language in file names, headers, tags, and body content.

## Templates

Override default note templates for specific file types.

Example:

## Templates

### Journal Override
My daily journal should have these sections:

{Project/Topic}

  • What happened
  • Decisions made

Reflections

  • What went well
  • What to improve

### People Override
Add a "## Shared Interests" section after "## Who" in new People files.

How it works: When creating new files, Claude checks this section first. If a template override exists for the file type, it uses your template instead of the default from assets/templates/.

## Notes

Free-form space for anything else you want the brain to know.

Example:

## Notes

- I'm a morning person — weight morning captures higher than evening ones
- I use Obsidian daily, so wikilinks and graph compatibility matter a lot
- When in doubt about where something goes, ask me rather than guessing wrong

Cascading: Global + Project

Remember supports two levels of REMEMBER.md that layer together:

Global REMEMBER.md

Lives at your brain root (e.g., ~/remember/REMEMBER.md). Contains your universal preferences that apply everywhere — capture rules, language settings, processing defaults, etc.

Project REMEMBER.md

Lives at any project root (e.g., ./REMEMBER.md in your current working directory). Contains project-specific additions that layer on top of global preferences.

How Merging Works

Project sections APPEND to global sections — they don't replace them.

If both files have a ## Capture Rules section, the content is concatenated. Your global rules still apply, and the project adds specificity on top.

Example:

Global ~/remember/REMEMBER.md:

## Capture Rules

### Always Capture
- Decisions with rationale
- Task assignments

### Never Capture
- Debugging sessions
- Casual greetings

## Language
- Default: English for headers and tags
- Romanian OK for content

Project ./REMEMBER.md:

## Capture Rules

### Always Capture
- API contract changes (this project has strict API versioning)
- Performance benchmarks

### Never Capture
- CI/CD pipeline noise

## Custom Types

### API Change Log (→ Notes/api-changes/)
Track breaking changes with: endpoint, version, migration notes

Merged result (what Claude sees):

  • Capture Rules = global rules + project rules (all apply)
  • Language = global only (project didn't override)
  • Custom Types = project only (global had none)

When to Use Project REMEMBER.md

  • Project has unique routing needs (e.g., custom entity types)
  • Different capture thresholds than your global defaults
  • Project-specific "always capture" or "never capture" rules
  • Team conventions that differ from your personal preferences

Getting Started

After running /remember:init, you'll find a starter REMEMBER.md with empty sections:

# REMEMBER.md

Instructions for how your Second Brain captures and processes knowledge.
All sections are optional.

---

## Capture Rules

## Processing

## Custom Types

## Connections

## Language

## Templates

## Notes

Just fill in the sections that matter to you. Empty sections are ignored — defaults apply.

Tips

  1. Start small — add one or two Capture Rules, see how it works, then expand
  2. Be specific — "Never capture debugging sessions" is clearer than "Don't save technical stuff"
  3. Use examples — "Route like this: X → Y" is better than abstract descriptions
  4. Review periodically — as your workflow evolves, update REMEMBER.md to match
  5. Don't duplicate Persona.md — REMEMBER.md is for instructions, Persona.md is for patterns. Let each do its job.

FAQ

Q: What happens if REMEMBER.md conflicts with Persona.md? A: REMEMBER.md wins. It represents your explicit intent, which takes precedence over observed patterns.

Q: Can I delete sections I don't use? A: Yes. Missing sections use defaults. You can also leave them empty — same effect.

Q: Will /remember:process modify my REMEMBER.md? A: Never. REMEMBER.md is exclusively user-edited. Only Persona.md is auto-updated.

Q: How many tokens does REMEMBER.md use? A: A moderate file (~50 lines) uses ~500 tokens. For brain dumps, only Capture Rules, Processing, Custom Types, and Language sections are injected (not the full file). For /remember:process, the full file is loaded (batch operation with larger token budget).

Q: Can I have project-specific REMEMBER.md files? A: Yes! Place a REMEMBER.md in your project root. Its sections append to your global REMEMBER.md at the brain root. See Cascading: Global + Project above.