| name | reflection-global |
| description | Analyze conversation to improve global ~/CLAUDE.md with cross-project preferences and behavior patterns. Use this when Claude repeated a mistake it makes everywhere, when you had to re-explain a preference that should apply to all projects, or when you notice a communication or workflow pattern worth capturing globally. Trigger with "reflect globally", "update global instructions", "capture cross-project preference", or "improve global CLAUDE.md". Also use when /reflection surfaces findings that aren't project-specific. |
Global CLAUDE.md Reflection
Analyze this conversation to improve ~/CLAUDE.md for future sessions across all projects.
Scope
This skill targets global, cross-project behavior — how Claude communicates, preferred tools, workflow patterns, and user preferences that apply everywhere.
- If an issue is specific to a single project's conventions, architecture, or tooling, use
/reflection instead to update that project's CLAUDE.md.
- If a finding is about user preferences or feedback about Claude's behavior style, consider whether it belongs in the memory system (
~/.claude/projects/*/memory/) rather than ~/CLAUDE.md. ~/CLAUDE.md is for durable instructions; memory is for contextual preferences and behavioral feedback.
When to Use
- Claude misunderstood a request in this session
- A preference or pattern had to be corrected or explained
- A recurring workflow isn't captured in global instructions
- Claude's behavior was inconsistent with stated preferences across projects
- Something worked unexpectedly well and should be reinforced globally
Workflow
Phase 1: Analysis
Review the conversation history first, then read ~/CLAUDE.md to identify what's missing or inconsistent.
Look for moments where:
- Misunderstandings occurred - You interpreted instructions incorrectly
- Preferences were missing - User had to re-explain a cross-project preference
- Patterns were unclear - You had to ask or guess about conventions that apply globally
- Workflow friction - Steps that could be streamlined with better global guidance
- Repeated corrections - The user corrected you multiple times on the same issue
- What worked well - A non-obvious approach was validated; worth reinforcing globally
For each finding, determine:
- What happened: Specific moment in the conversation
- Root cause: What was missing or unclear in
~/CLAUDE.md
- Proposed fix: Exact text to add or modify
- Destination:
~/CLAUDE.md (global instructions) or memory system (behavioral feedback)
Phase 2: Interaction
Present all findings grouped by destination. Wait for user approval per finding.
If a proposed change is rejected, refine or move on.
Only propose improvements when there's clear evidence of recurring issues or missing context — not every session needs changes.
Phase 3: Implementation
For each approved change:
- Edit
~/CLAUDE.md directly using the Edit tool
- For findings destined for memory, write the appropriate memory file under
~/.claude/projects/-Users-<username>--claude/memory/
Common Mistakes
- Proposing changes based on a single edge case — look for patterns across the conversation
- Rewriting sections that are working well — surgical edits only
- Adding rules that duplicate existing ones — check for overlap before adding
- Putting project-specific conventions here — those belong in the project's
CLAUDE.md
- Putting behavioral feedback in
~/CLAUDE.md when it belongs in memory
Output Format
## Reflection Summary
### Issue 1: [Brief title]
**What happened:**
> [Quote or describe the specific conversation moment]
**Root cause:**
The ~/CLAUDE.md doesn't specify [what was missing]
**Proposed addition to ~/CLAUDE.md:**
Add to section `[Section Name]`:
> [Exact text to add]
**Destination:** ~/CLAUDE.md / memory
---
### Issue 2: [Brief title]
[...]
If No Issues Found
Say so briefly. Not every session needs ~/CLAUDE.md changes.