>> claude-memory-user-edits-guide
stars: 2
forks: 0
watches: 2
last updated: 2025-12-10 10:56:54
Memory User Edits: Expert Guidance
THE CORE INSIGHT
Memory edits work for FACTS, not BEHAVIORS.
| ✅ Works (Facts) | ❌ Doesn't Work (Behaviors) |
|---|---|
| "Backend stores 7KB per user" | "Always check docs first" |
| "Architecture: two-layer model" | "Prioritize architecture over details" |
| "User prefers TypeScript" | "Never suggest JavaScript" |
THE FOUR COMMANDS
| Command | Purpose | Parameters |
|---|---|---|
view | See current edits | None |
add | Create new edit | control (max 200 chars) |
remove | Delete by line number | line_number |
replace | Update existing | line_number, replacement |
Constraints:
- Maximum 30 edits total
- 200 characters per edit
- Project-scoped only (not regular chats)
- Plain text only (no markdown/HTML)
WHEN TO USE memory_user_edits
Use for:
- User says "remember that...", "don't forget...", "please note..."
- User asks "what do you know about me?"
- User wants to correct a fact Claude keeps getting wrong
- User shares persistent preferences or constraints
Don't use for:
- Behavioral instructions (use Project Instructions instead)
- Temporary information
- Complex conditional logic
- Information that changes frequently
BEST PRACTICES
Writing Good Edits
✅ Good (atomic, factual):
"Backend: PostgreSQL 15 with Knex.js" (36 chars)
"User timezone: CET (UTC+1)" (26 chars)
"Project deadline: January 2026" (30 chars)
❌ Bad (vague, behavioral):
"Uses good database practices" (vague)
"Always check architecture first" (behavioral)
"Remember to prioritize user needs" (instruction)
Priority Framework
When approaching 30 edits, prioritize:
| Priority | Type | Action |
|---|---|---|
| P0 | Facts Claude gets wrong repeatedly | Keep always |
| P1 | Important preferences, key definitions | Keep if space |
| P2 | Nice-to-haves, well-documented info | Remove first |
RESPONSE PROTOCOL
When user asks about memory:
- View first: Always check current edits before adding
- Check for duplicates: Don't add redundant information
- Verify facts: Ensure it's a FACT, not a BEHAVIOR
- Confirm action: Tell user what was added/removed/changed
When user says "remember X":
- Rewrite X as a concise fact (under 200 chars)
- Check for conflicts with existing edits
- Add the edit
- Confirm: "Added memory #N: [content]"
TROUBLESHOOTING
"Claude ignores my memory edits"
- Verify you're in a Claude Project (not regular chat)
- Check if edit is a FACT (works) vs BEHAVIOR (doesn't work)
- Ensure query is relevant to the stored fact
"Memory edits work inconsistently"
- Edit may be too vague → make more specific
- Documentation may contradict → align them
- Edit may be behavioral → rewrite as fact
"Hit 30-edit limit"
- View all edits
- Remove P2 (nice-to-haves) first
- Consolidate related edits
QUICK REFERENCE
Add a fact:
add control="[concise fact under 200 chars]"
View all edits:
view
Remove an edit:
remove line_number=[N]
Update an edit:
replace line_number=[N] replacement="[new text]"
THE FOUR-LAYER SYSTEM
Memory edits work best as part of a complete system:
| Layer | Purpose |
|---|---|
| Project Instructions | HOW Claude should work (behaviors) |
| Documentation | Detailed project knowledge |
| Memory Edits | Critical facts always in focus |
| Active Steering | Real-time corrections |
Key: Memory edits amplify Project Instructions—they don't replace them.
Memory User Edits Guide by Francesco Marinoni Moretto — CC BY 4.0
