Feature Request: Session Templates / Forking - Save compressed state as starting point for new sessions

Feature Request: Session Templates / Forking

Summary

Ability to save a session's compressed context state as a template, then spawn multiple new sessions that start from that frozen state (not resuming the original session).

Problem Statement

Current Limitation:

• Each new session starts with minimal context
• Must re-read files, re-learn project structure every time
• Cannot share "rich context" across multiple parallel sessions
• Session resume only allows ONE continuation, not forking

Real-World Pain Point:

Session 1: Read 50 files → /compact → [Rich Context]
Now I want:
  - Session 2: Start from Session 1's state (work on feature A)
  - Session 3: Start from Session 1's state (work on feature B)
  - Session 4: Start from Session 1's state (fix bug)
  
Currently: IMPOSSIBLE
Each session must re-read files = wasted time + tokens

Proposed Solution

CLI Commands

# Save current session as template
/session save --name "sprint-5-context" --description "Sprint 5.2 with all domain rules read"

# List available templates
/session list --templates

# Create NEW session from template (NOT resume)
/session new --template "sprint-5-context" --name "Implementing US-53"

# Delete old template
/session delete-template "sprint-5-context"

# Auto-save after file-reading heavy sessions
/session save --name "backend-context" --if-files-read-greater 20

UI/UX

In VS Code extension:

• Right-click session → "Save as Template"
• New Session → "Start from Template" dropdown
• Template manager (view/delete/update)

In CLI:

• Template names with tab completion
• Show template size/age
• Validate template integrity

Use Cases

1. Project Onboarding

# Senior dev: Read entire codebase once
Session 1: Read 100 files → /compact → /session save --name "njordbok-full"

# Junior devs: Start from that context
Session 2: /session new --template "njordbok-full" --name "Feature X"
Session 3: /session new --template "njordbok-full" --name "Bug Y"

2. Parallel Feature Development

# After sprint planning meeting
Session 1: Read all sprint docs + domain rules → /session save --name "sprint-6-kickoff"

# Parallel work
Session 2: /session new --template "sprint-6-kickoff" # US-101
Session 3: /session new --template "sprint-6-kickoff" # US-102
Session 4: /session new --template "sprint-6-kickoff" # US-103

3. Knowledge Reuse

# Deep dive into auth system once
Session: Read auth/ → /session save --name "auth-expert-context"

# Reuse for all auth-related work
Session: /session new --template "auth-expert-context" # Work on auth bug
Session: /session new --template "auth-expert-context" # Work on RBAC feature

4. Team Collaboration

# Tech lead creates template
/session save --name "team-shared-context" --description "Q2 architecture decisions"

# Team members use it
/session new --template "team-shared-context" # Everyone starts aligned

Technical Considerations

Storage Location

• Local: .claude/templates/ (project-specific)
• Global: ~/.claude/templates/ (shared across projects)
• Cloud: (future) Sync templates across devices

Template Contents

Should include:

• ✅ Compressed message history
• ✅ Files read (with checksums for validation)
• ✅ Learnings/context from conversation
• ❌ Sensitive data (API keys, secrets - auto-filter)
• ❌ Task state (start fresh for new sessions)

Validation

# Check if template is still valid
/session validate-template "sprint-5-context"
# Warns if:
# - Files have changed (git diff)
# - Project structure changed
# - Template is > 30 days old

Size Management

# Show template sizes
/session list --templates --sizes

# Auto-cleanup
/session cleanup --templates --older-than 90days

Alternative Names

• Session Templates
• Session Forking
• Session Snapshots
• Context Templates
• Saved States

Priority

High - Would significantly improve workflow for:

• Large codebases (read files once, reuse everywhere)
• Teams (shared starting points)
• Parallel work (multiple sessions from same context)
• Token efficiency (avoid re-reading same files)

Related Features

• Memory files (auto-loaded, but manual)
• Session resuming (continuation, not forking)

Implementation Ideas

Option 1: Full Context Clone

• Save entire compressed conversation
• New sessions load this as initial messages
• Simpler but larger storage

Option 2: Structured Snapshot

• Extract key learnings, files, context
• Store as structured JSON
• More efficient, but complex extraction

Option 3: Hybrid

• Save summary + file list + key decisions
• New sessions read files from list (fast, current)
• Best balance of size/accuracy

---

Would greatly improve productivity for teams working with large codebases!