## Critical Fixes ✅ - **CLI Command Syntax**: Fixed all ambiguous `SuperClaude --version` → `python3 -m SuperClaude --version` - **Architecture Clarity**: Verified dual architecture documentation (Python CLI + Context Framework) - **External Dependencies**: Marked unverified APIs as experimental (TWENTYFIRST_API_KEY, MORPH_API_KEY) - **Installation Instructions**: Clarified NPM package names with verification warnings ## Content Optimization 🗑️ - **Removed unnecessary files**: - optimization-guide.md (inappropriate for context files) - quick-start-practices.md (duplicate content) - Various outdated socratic learning components - **Updated cross-references**: Fixed all broken links to point to existing, relevant content - **Consolidated navigation**: Streamlined Reference/README.md documentation matrix ## Technical Accuracy 🎯 - **Command References**: All commands now specify exact usage context (terminal vs Claude Code) - **Framework Nature**: Consistently explains SuperClaude as context framework, not executable software - **Installation Verification**: Updated diagnostic scripts with correct Python CLI commands - **MCP Configuration**: Marked experimental services appropriately ## Impact Summary 📊 - **Files Modified**: 15+ documentation files for accuracy and consistency - **Files Removed**: 5+ unnecessary/duplicate files - **Broken Links**: 0 (all cross-references updated) - **User Clarity**: Significantly improved understanding of dual architecture Result: Professional documentation that accurately represents SuperClaude's sophisticated dual architecture (Python CLI installation system + Claude Code context framework). 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
6.4 KiB
SuperClaude Session Management Guide
✅ Important Note
Sessions are conversation-scoped, not persistent between Claude Code conversations.
Understanding Session Context
What "Sessions" Actually Are
Reality Check: Claude Code conversations are independent. Each new conversation starts fresh. "Session management" in SuperClaude refers to commands that help Claude understand and work with your project context during a single conversation.
What /sc:load and /sc:save Actually Do
/sc:load: Tells Claude to analyze and understand a project structure within the current conversation/sc:save: Creates a summary or checkpoint of work done in the current conversation/sc:reflect: Reviews what has been discussed/done in the current conversation
Important: These commands DO NOT persist data between Claude Code conversations. Each conversation is isolated.
How Context Works in Claude Code
Within a Conversation
During a single Claude Code conversation:
- Claude maintains context of what you've discussed
- Can refer back to earlier parts of the conversation
- Understands the project structure you've shown
- Remembers decisions made in this conversation
Between Conversations
When you start a new Claude Code conversation:
- Previous conversation context is lost
- Must re-explain project context
- Previous decisions aren't remembered
- Must reload any necessary understanding
Session Commands Explained
/sc:load - Project Context Loading
What it does: Analyzes project structure and creates mental model What it doesn't do: Store anything persistently
/sc:load src/
# Claude analyzes the src/ directory structure
# Builds understanding for THIS conversation only
Use cases:
- Starting work on an existing project
- Helping Claude understand codebase structure
- Establishing context for subsequent commands
/sc:save - Conversation Summary
What it does: Creates a summary of work done What it doesn't do: Actually save for next conversation
/sc:save "refactoring-complete"
# Claude summarizes what was accomplished
# Useful for you to copy/paste to notes
# Does NOT persist for next conversation
Use cases:
- Creating documentation of work done
- Generating summary for your records
- Checkpoint before major changes
/sc:reflect - Conversation Review
What it does: Reviews current conversation progress What it doesn't do: Access previous conversations
/sc:reflect
# Claude reviews what's been discussed
# Summarizes decisions and progress
# Only covers current conversation
Use cases:
- Mid-conversation status check
- Reviewing decisions made
- Planning next steps
Working Effectively Without Persistence
Starting New Conversations
When starting a new Claude Code conversation for continued work:
-
Provide Context Again
/sc:load project-directory/ # Re-establishes project understanding -
Summarize Previous Work
"We previously refactored the auth system. Now we need to..." -
Reference Key Decisions
"We're using JWT tokens with refresh tokens, as decided earlier"
Creating Your Own Persistence
Since Claude doesn't persist between conversations, create your own:
-
Project Notes File
# project-notes.md ## Session 1 (Date) - Implemented user authentication - Decided on JWT with refresh tokens - Created user model with email validation ## Session 2 (Date) - Added password reset functionality - Integrated email service -
Copy Important Summaries
/sc:save "session-end" # Copy the output to your notes -
Document Key Decisions
/sc:reflect # Copy important decisions to your documentation
MCP Servers and "Memory"
Serena MCP Server
If Serena MCP is configured, it may provide some project analysis capabilities, but:
- Still doesn't persist between Claude conversations
- Operates within MCP server constraints
- Not true persistent memory
What MCP Servers Can Do
- Provide tools for file analysis
- Help with project navigation
- Enhance Claude's capabilities
- But NOT store conversation history
Best Practices
For Single Conversation
-
Load Context Early
/sc:load . # Start with full project context -
Work Systematically
/sc:workflow "plan the session" # Organize your work -
Save Summaries
/sc:save "checkpoint-1" # Create summaries as you go
For Multiple Conversations
-
Maintain External Notes
- Keep a project journal
- Document decisions
- Track progress
-
Create Context Files
# .claude-context.md Project context for Claude Code sessions - Architecture: microservices - Database: PostgreSQL - Auth: JWT with refresh tokens -
Start Consistently
# At start of each conversation: /sc:load . "Please read .claude-context.md for project decisions"
Common Misconceptions
❌ Myth: Sessions Persist
Reality: Each Claude conversation is independent
❌ Myth: /sc:save Stores Data
Reality: It creates a summary for you to copy
❌ Myth: /sc:load Remembers Previous Work
Reality: It analyzes current files only
❌ Myth: Serena Provides Memory
Reality: MCP tools work within conversation scope
Troubleshooting
"Claude doesn't remember our previous conversation"
Expected behavior: Claude Code conversations don't persist Solution: Provide context at start of new conversation
"/sc:load isn't loading previous session"
Expected behavior: It loads project files, not session history Solution: Maintain your own project notes
"/sc:save doesn't actually save"
Expected behavior: Creates summary text only Solution: Copy output to your own notes
Summary
SuperClaude's "session management" commands are tools for managing context within a single Claude Code conversation. They help Claude understand your project and track progress during the current conversation, but do not provide persistence between conversations.
For true continuity across multiple conversations, maintain your own documentation and provide context at the start of each new conversation. Think of these commands as conversation helpers, not database operations.