Baton - Context Management Skill

Manages context across sessions and compactions using TLDR summaries and structured documentation.

Implementation

When

/baton

Core Commands

• No arguments or

  load

  : Display current TLDR summary

• init

  : Initialize new conversation

• rename <title>

  : Set conversation title (max 60 chars)

• rename --suggest

  : Get AI-generated title suggestions

• save [note]

  : Save current state with optional note

• update [section]

  : Update SUMMARY.md sections automatically

• update --auto

  : Auto-generate updates from recent activity

• history

  : Show all conversations with titles

• status

  : Show token usage and save recommendation

• archive

  : Archive completed items to prevent file bloat

Navigation & Switching

• switch <conv-id-or-title>

  : Switch to different conversation

• switch --recent

  : Show recent conversations to choose from

Search & Discovery

• search <term>

  : Search across all conversations, bugs, decisions

• search --bugs <term>

  : Search only bugs

• search --decisions <term>

  : Search only decisions

• search --conversations <term>

  : Search only conversations

• context <topic>

  : Load relevant past context (smart search)

Reporting & Analytics

• report [timeframe]

  : Generate work summary report

• report --today

  : Today's work summary

• report --week

  : Past week summary

• report --conversation <id>

  : Specific conversation report

• metrics

  : Show baton system effectiveness metrics

• stats

  : Alias for metrics

Validation & Health

• validate

  : Check file structure integrity

• health

  : Alias for validate

Configuration

• auto-save on|off|status

  : Configure auto-save triggers

• template create <name>

  : Create custom SUMMARY.md template

• template use <name>

  : Switch to custom template

• template list

  : Show available templates

Git Integration

• git-link

  : Associate conversation with current git branch

• git-summary

  : Generate commit message from SUMMARY.md

Implementation Details

Each command should be implemented by reading/writing the appropriate files in

.claude/

For

/baton init

CONV_ID="conv-$(date +%Y%m%d-%H%M%S)"
mkdir -p .claude/conversations/$CONV_ID
echo $CONV_ID > .claude/CURRENT_CONVERSATION_ID
# Create SUMMARY.md from template
# Initialize shared files if they don't exist
touch .claude/{BUGS.md,DECISIONS.md,CONVERSATION_HISTORY.md,ENHANCEMENTS.md,USER_FEEDBACK.md}
mkdir -p .claude/{archive,templates}
# Create default settings if missing
cat > .claude/settings.json <<EOF
{
  "autoSave": {
    "enabled": true,
    "thresholds": [70, 85, 95],
    "notifyOnSave": true
  },
  "archiveThresholds": {
    "conversations": 10,
    "bugs": 20,
    "decisions": 15
  },
  "defaultTemplate": "standard"
}
EOF

For

/baton load

• Read

  .claude/CURRENT_CONVERSATION_ID

• Display

  .claude/conversations/{conv-id}/SUMMARY.md

• Display filtered entries from

  .claude/BUGS.md

  (matching conversation ID)
• Display filtered entries from

  .claude/DECISIONS.md

  (matching conversation ID)
• Display current TODO list if available

For

/baton rename <title>

• Read

  .claude/CURRENT_CONVERSATION_ID

• Update

  .claude/conversations/{conv-id}/SUMMARY.md

  Title field
• Update

  .claude/CONVERSATION_HISTORY.md

  entry for this conversation
• Validate title length (max 60 chars for clean display)
• Used in standardized response format header
• Example:

  /baton rename Self-Improving AI Chatbot

For

/baton rename --suggest

• Analyze recent work: TODO items, file changes, recent conversation
• Generate 3-5 concise title suggestions (max 60 chars each)
• Display suggestions with numbers for easy selection
• Allow user to choose or provide custom title
• Example output:

  Suggested titles based on recent work:
  1. API Authentication System
  2. User Login Bug Fixes
  3. Database Schema Migration
  4. Payment Integration Setup
  5. Custom: [Enter your own]
  
  Choose (1-5) or press Enter for custom:
  

For

/baton save [note]

• Read current state (TODO list, recent changes)
• Update

  .claude/conversations/{conv-id}/SUMMARY.md

• Add timestamp and optional note
• Update Task Checklist, Key Files, State Snapshot sections
• Show confirmation: "Saved at 45% token usage"

For

/baton update [section]

Auto-update specific sections:

• --auto

  : Analyze recent activity and update all sections

• tasks

  : Update Task Checklist from TODO list

• files

  : Update Key Files from recent git changes

• state

  : Update State Snapshot from current file/line

• context

  : Update Context in 3 Lines from recent work

Implementation:

• Read recent TODO completions, file modifications
• Analyze conversation for decisions, bugs discovered
• Update relevant SUMMARY.md sections
• Show what was updated: "Updated: Task Checklist (3 completed), Key Files (2 modified)"

For

/baton history

• Display

  .claude/CONVERSATION_HISTORY.md

• Format with conversation IDs, titles, status, dates
• Show active vs completed conversations

For

/baton status

• Check current token usage (estimate from conversation length)
• Calculate percentage to 70% threshold
• Show time since last SUMMARY.md update
• Recommend save if > 70%
• Check file sizes and recommend archival if thresholds exceeded:
  • CONVERSATION_HISTORY.md > 10 conversations
  • BUGS.md > 20 bugs
  • DECISIONS.md > 15 decisions
• Display auto-save configuration status

For

/baton archive

• Count items in each file by status
• Move completed/fixed/implemented items to

  .claude/archive/

• Archive structure:

  • archive/conversations/YYYY-MM.md

    (completed conversations by month)

  • archive/bugs/fixed-YYYY-MM.md

    (fixed bugs by month)

  • archive/decisions/implemented-YYYY-MM.md

    (implemented decisions by month)
• Update main files to remove archived items
• Show summary: "Archived 5 conversations, 12 bugs, 8 decisions"
• Keep last 10 active conversations, all active/in-progress bugs, all proposed/accepted decisions

For

/baton switch <conv-id-or-title>

Switch between conversations:

• Accept either conversation ID or title (fuzzy match)
• Update

  .claude/CURRENT_CONVERSATION_ID

• Load new conversation's SUMMARY.md
• Show transition message: "Switched from 'API Redesign' → 'Bug Fixes'"
• Display new conversation's current state

With

--recent

• List 5-10 most recent conversations
• Show: ID, Title, Last Modified, Status
• Allow numbered selection
• Example:

  Recent conversations:
  1. [Active] API Authentication System (modified 2h ago)
  2. [Active] Database Migration (modified 5h ago)
  3. [Paused] Payment Integration (modified 2d ago)
  
  Choose conversation (1-3):
  

For

/baton search <term>

Search across all baton files:

• Search SUMMARY.md files in all conversations
• Search BUGS.md, DECISIONS.md, ENHANCEMENTS.md, USER_FEEDBACK.md
• Return matches with context (file, conversation ID, line)
• Highlight matching text
• Sort by relevance (exact match > partial match)

Example output:

Found 'authentication' in 5 locations:

CONVERSATIONS:
  conv-20251224-015859: API Authentication System
    - Line 12: "Implementing OAuth2 authentication"

  conv-20251223-140521: User Login Refactor
    - Line 8: "Fixed authentication token expiry bug"

DECISIONS:
  Decision #3 (conv-20251224-015859)
    - "Chose OAuth2 over email/password authentication"

BUGS:
  Bug #7 (conv-20251223-140521)
    - "Authentication fails after 24 hours"

With specific filters:

• --bugs

  : Search only BUGS.md

• --decisions

  : Search only DECISIONS.md

• --conversations

  : Search only SUMMARY.md files

For

/baton context <topic>

Smart context loading:

• Semantic search across past conversations for topic
• Load relevant decisions, bugs, failed attempts
• Show context from most relevant conversations
• Example:

  /baton context authentication

Output:

Loading context for 'authentication'...

Found 3 relevant conversations:

1. conv-20251224-015859: API Authentication System (Active)
   Decisions: OAuth2 implementation, JWT tokens
   Bugs: None currently
   Status: 85% complete

2. conv-20251223-140521: User Login Refactor (Completed)
   Decisions: Session timeout 24h
   Bugs: Fixed token expiry issue
   Failed: Email/password approach (too complex)

3. conv-20251220-093012: Security Audit (Completed)
   Decisions: Require 2FA for admin accounts

Key learnings:
- Avoid email/password (maintenance overhead)
- Use OAuth2 with established providers
- JWT tokens preferred over sessions

For

/baton report [timeframe]

Generate work summary reports:

• Aggregate work across conversations
• Count tasks completed, bugs fixed, decisions made
• List key files modified
• Perfect for standups, status updates

Timeframes:

• --today

  : Today's work

• --week

  : Past 7 days

• --month

  : Past 30 days

• --conversation <id>

  : Specific conversation
• No flag: Ask for timeframe

Example output:

## Work Summary: Past 7 Days (Dec 18-25, 2025)

**Active Conversations:** 3
**Completed Conversations:** 2

### Tasks Completed: 24
- API authentication system (12 tasks)
- Database migration (8 tasks)
- Bug fixes (4 tasks)

### Bugs Fixed: 5
- Authentication token expiry
- Database connection timeout
- API rate limiting error
- UI rendering glitch
- Memory leak in cron job

### Decisions Made: 8
- OAuth2 for authentication
- PostgreSQL for main database
- Redis for caching
- Next.js for frontend
- ... (4 more)

### Files Modified: 47
Top 10:
- src/auth/oauth.ts (23 changes)
- src/db/schema.sql (18 changes)
- src/api/endpoints.ts (15 changes)
... (7 more)

### Key Achievements:
- Completed API authentication system
- Migrated database schema to v2
- Reduced bug count from 12 → 7

For

/baton metrics

or

/baton stats

Show system effectiveness:

• Total conversations tracked
• Token savings vs full conversation logs
• Compaction survival rate
• Average context restoration time
• File sizes and growth trends

Example output:

📊 Baton System Metrics

CONVERSATIONS:
  Total: 12 (3 active, 9 completed)
  Avg Duration: 3.2 hours
  Avg Compactions: 4.2 per conversation
  Longest: conv-20251215-081234 (12 compactions)

TOKEN EFFICIENCY:
  Full Conversation Logs: ~612,000 tokens
  TLDR Summaries: ~6,200 tokens
  Compression Ratio: 98.9% (99x reduction)
  Estimated Cost Savings: $18.36 (at $0.03/1K tokens)

CONTEXT RESTORATION:
  Success Rate: 100% (12/12 compactions survived)
  Avg Restoration Time: 1.2 seconds
  Avg Tokens Read: 1,150 per restoration

FILE HEALTH:
  CONVERSATION_HISTORY.md: 2.1 KB (12 entries)
  BUGS.md: 1.8 KB (7 active, 15 archived)
  DECISIONS.md: 2.4 KB (11 active, 8 implemented)
  Archive Size: 18.2 KB (47 archived items)

RECOMMENDATIONS:
  ✅ System healthy
  ℹ️ Consider archiving (BUGS.md has 22 total entries)

For

/baton validate

or

/baton health

Check file integrity:

• Verify

  .claude/

  directory structure exists
• Check all required files present
• Validate SUMMARY.md has required sections
• Detect missing conversation IDs, broken tags
• Check for orphaned files
• Validate settings.json syntax

Example output:

🔍 Baton Health Check

DIRECTORY STRUCTURE:
  ✅ .claude/ exists
  ✅ .claude/conversations/ exists
  ✅ .claude/archive/ exists
  ✅ .claude/templates/ exists

REQUIRED FILES:
  ✅ CONVERSATION_HISTORY.md (2.1 KB)
  ✅ BUGS.md (1.8 KB)
  ✅ DECISIONS.md (2.4 KB)
  ✅ CURRENT_CONVERSATION_ID (present)
  ✅ settings.json (valid JSON)
  ⚠️ ENHANCEMENTS.md (empty)
  ⚠️ USER_FEEDBACK.md (empty)

CONVERSATIONS:
  ✅ conv-20251224-015859: All sections present
  ✅ conv-20251223-140521: All sections present
  ❌ conv-20251220-093012: Missing "Failed Attempts" section

TAGS:
  ✅ All BUGS.md entries have Conv: tags
  ⚠️ 2 DECISIONS.md entries missing Conv: tags

ISSUES FOUND: 3
  [1] conv-20251220-093012/SUMMARY.md missing "Failed Attempts"
  [2] DECISIONS.md line 45 missing Conv: tag
  [3] DECISIONS.md line 67 missing Conv: tag

FIX COMMAND:
  /baton fix --auto   # Auto-fix common issues

For

/baton auto-save on|off|status

Configure automatic saves:

• on

  : Enable auto-save at configured thresholds

• off

  : Disable auto-save

• status

  : Show current configuration

Auto-save behavior:

• Monitor token usage continuously
• Save at thresholds: 70%, 85%, 95% (configurable)
• Show notification: "🔁 Auto-saved at 72% token usage"
• Never interrupt user work
• Settings stored in

  .claude/settings.json

Example output:

Auto-Save Configuration:
  Status: ✅ Enabled
  Thresholds: 70%, 85%, 95%
  Notify on save: Yes
  Last auto-save: 15 minutes ago (at 71%)

Current token usage: 68% (save at 70%)

For

/baton template create <name>

Create custom templates:

• Copy current SUMMARY.md structure
• Save to

  .claude/templates/<name>.md

• Allow editing/customization
• Templates stored for reuse

Example:

/baton template create research

Created template: research
Location: .claude/templates/research.md

Customize with additional sections:
- Research Questions
- Literature Review
- Methodology
- Data Sources

For

/baton template use <name>

Switch templates:

• Load template from

  .claude/templates/<name>.md

• Apply to new conversations
• Existing conversations keep their template

For

/baton template list

Show available templates:

Available Templates:
  ✓ standard (default) - Standard development template
  • research - Academic research template
  • devops - Infrastructure/deployment template
  • bugfix - Bug investigation template

Current conversation template: standard

For

/baton git-link

Git integration:

• Associate current conversation with git branch
• Store mapping in

  .claude/git_links.json

• Add conversation ID to commit template
• Tag commits with conversation context

Example:

/baton git-link

✅ Linked conversation to git:
   Conv: conv-20251224-015859
   Branch: feature/oauth-authentication

Commits on this branch will be tagged with conversation ID.
Use /baton git-summary to generate commit message.

For

/baton git-summary

Generate commit message from SUMMARY:

• Extract Task Checklist completed items
• Summarize key changes
• Include conversation ID for traceability

Example output:

feat: implement OAuth2 authentication system

- Add OAuth2 provider integration (Google, GitHub)
- Create JWT token generation and validation
- Implement user session management
- Add authentication middleware
- Write integration tests for auth flow

Related to conversation: conv-20251224-015859

Problem Solved

During long autonomous sessions, Claude Code can go through 5-10 auto-compactions, losing critical technical details:

• Bug reproduction steps
• Failed approaches (leads to retry loops)
• Architecture decisions and rationale
• Exact file locations and current state
• What was tried and why it didn't work

Solution

Two-tier context system with intelligent automation:

• Tier 1 (TLDR): ~1,000 tokens - Always read after compaction
• Tier 2 (Full details): ~50,000 tokens - Read on-demand only
• Compression: 25-100x token reduction
• Auto-save: Never lose work at compaction boundaries
• Search: Find past context instantly
• Smart updates: Reduce manual overhead

Commands Reference

Core Commands

/baton init

/baton load

/baton

/baton save [note]

/baton update [section]

/baton rename <title>

/baton rename --suggest

/baton history

/baton status

/baton archive

Navigation

/baton switch <conv-id-or-title>

/baton switch --recent

Search & Discovery

/baton search <term>

/baton search --bugs <term>

/baton search --decisions <term>

/baton context <topic>

Reporting

/baton report [timeframe]

/baton metrics

/baton stats

Validation

/baton validate

/baton health

Configuration

/baton auto-save on|off|status

/baton template create <name>

/baton template use <name>

/baton template list

Git Integration

/baton git-link

/baton git-summary

ENHANCEMENTS.md - Future Ideas Tracking

Track potential improvements and feature ideas:

## Enhancement #1: Add Dark Mode
**Conv:** conv-20251224-015859
**Proposed:** 2025-12-24
**Priority:** High | Medium | Low
**Status:** Proposed | Accepted | In Progress | Implemented | Rejected
**Rationale:** Users requested dark mode for nighttime viewing
**Impact:** Affects all UI components, requires theme system
**Effort:** ~2 days
**Dependencies:** None
**Proposed By:** User | Claude
**Notes:** Consider CSS variables for easy theming

When to add:

• User mentions "we should..." or "it would be nice if..."
• Claude identifies optimization opportunities
• Discussing future improvements
• Brainstorming features

Status progression:

• Proposed → Accepted → In Progress → Implemented
• Or: Proposed → Rejected (with reason)

USER_FEEDBACK.md - Questions Waiting for User

Critical for long autonomous sessions when user is away:

## Feedback Request #1: Authentication Method
**Conv:** conv-20251224-015859
**Asked:** 2025-12-24 03:15
**Risk Level:** High | Medium | Low
**Status:** Pending | Auto-Decided | Answered | No Longer Needed
**Context:** Building user authentication system
**Question:** Should we use OAuth2 or email/password auth?
**Options:**
1. OAuth2 (Google + GitHub) - Easier for users, harder to implement
2. Email/Password - Traditional, requires password reset flow
3. Both - Best UX, most complex
**Blockers:** Can't proceed with auth implementation until decided
**Workaround:** Working on other features in the meantime
**Recommendation:** OAuth2 (most common for modern apps, aligns with best practices)
**Auto-Decision:** [If Low risk, Claude fills this in with choice made]
**User Validation:** [User confirms or requests redo]

Risk Assessment Guide:

High Risk - Block and wait for user (COSTS MONEY or irreversible):

• Costs money: Cloud provider choice (AWS vs GCP), paid services, API pricing
• Expensive to redo: 3+ days of rework if wrong choice
• Architecture decisions affecting entire system
• Breaking changes or data migration
• Security-critical choices (authentication, authorization, encryption)
• Irreversible decisions (data deletion, production deployments)
• User explicitly requested input on this type of decision
• RULE: If choosing wrong costs money or >2 days rework → HIGH RISK

Medium Risk - Auto-decide, document, validate (easy to redo):

• Framework/library choices (can swap in <1 day)
• UI/UX decisions (can redesign easily)
• Non-critical performance tradeoffs
• Feature prioritization
• RULE: If wrong choice costs <1 day to fix → MEDIUM RISK

Low Risk - Auto-decide, implement, validate later (trivial to redo):

• Variable naming conventions (<1 hour to refactor)
• File organization (can reorganize easily)
• Minor styling choices (quick CSS changes)
• Error message wording (find/replace)
• Default values (config change)
• Logging verbosity (config change)
• RULE: If wrong choice costs <1 hour to fix → LOW RISK

When to add:

• High Risk: Add to USER_FEEDBACK.md, block work on that feature
• Medium/Low Risk: Make decision, document in DECISIONS.md with status="Auto-Decided"

Auto-Decision Workflow (Low/Medium Risk):

1. Claude makes best-guess decision
2. Implements the feature
3. Documents in DECISIONS.md:

   **Status:** Auto-Decided (Pending User Validation)
   **Chosen:** OAuth2
   **Rationale:** Modern apps prefer OAuth, easier UX
   **Alternatives:** Email/password, Both
   **Risk:** Low - Can swap auth provider without data loss
   **Redo Effort:** ~2 hours if user prefers different approach
   

4. At next session start, ask user: "I chose OAuth2 for auth (modern standard). Alternatives were email/password or both. Okay with this?"
5. User validates (70% chance) or requests redo (30% chance)

Critical behavior:

• ALWAYS check USER_FEEDBACK.md at start of session
• Surface pending HIGH-RISK questions immediately
• For Auto-Decided items: Ask user for validation
• Be ready to redo work if user disagrees (~30% of time)
• Mark as "Validated" once user confirms

Auto-Behavior (No Command Needed)

Claude automatically follows these behaviors if

.claude/

On Session Start:

1. Check for

   .claude/CURRENT_CONVERSATION_ID

2. Read

   CONVERSATION_HISTORY.md

   (all conversations overview)
3. Read

   conversations/{conv-id}/SUMMARY.md

   (this conversation's TLDR)
4. Read

   BUGS.md

   and

   DECISIONS.md

   (filtered to this conversation)
5. Read

   USER_FEEDBACK.md

   - Surface any pending questions immediately
6. Read

   settings.json

   - Load auto-save configuration
7. Total: ~1,000-1,500 tokens for full context restoration

During Work:

• Use standardized response format (see CLAUDE.md) with Title, Request, Tasks, Summary
• Monitor token usage for auto-save triggers
• Update SUMMARY.md after significant actions
• Update conversation Title via

  /baton rename

  when conversation focus shifts
• Append to BUGS.md when discovering bugs
• Append to DECISIONS.md when making architecture choices
• Append to ENHANCEMENTS.md when ideas are discussed
• Append to USER_FEEDBACK.md when user input needed but user away
• Update CONVERSATION_HISTORY.md on major milestones
• Auto-save at 70%, 85%, 95% token thresholds if enabled

After Compaction:

• IMMEDIATELY run

  /baton load

  equivalent automatically
• Restore from TLDR (~1K tokens instead of 50K+ full log)
• Resume work with full context

Auto-Save Triggers:

• At 70% token usage: First checkpoint
• At 85% token usage: Second checkpoint
• At 95% token usage: Final checkpoint before compaction
• Show notification: "🔁 Auto-saved at 72% token usage"

File Structure

.claude/
├── CONVERSATION_HISTORY.md          # All conversations (~200 tokens)
├── BUGS.md                          # All bugs, tagged with conv-id
├── DECISIONS.md                     # All decisions, tagged with conv-id
├── ENHANCEMENTS.md                  # Future enhancement ideas
├── USER_FEEDBACK.md                 # Questions waiting for user input
├── CURRENT_CONVERSATION_ID          # Current conversation ID
├── settings.json                    # Baton configuration
├── git_links.json                   # Git branch associations
├── conversations/
│   └── {conv-id}/
│       └── SUMMARY.md               # This conversation TLDR (~300 tokens)
├── templates/
│   ├── standard.md                  # Default template
│   ├── research.md                  # Research project template
│   └── devops.md                    # DevOps/infrastructure template
└── archive/
    ├── conversations/
    │   └── YYYY-MM.md               # Archived conversations by month
    ├── bugs/
    │   └── fixed-YYYY-MM.md         # Fixed bugs by month
    └── decisions/
        └── implemented-YYYY-MM.md   # Implemented decisions by month

settings.json Configuration

{
  "autoSave": {
    "enabled": true,
    "thresholds": [70, 85, 95],
    "notifyOnSave": true
  },
  "archiveThresholds": {
    "conversations": 10,
    "bugs": 20,
    "decisions": 15
  },
  "defaultTemplate": "standard",
  "gitIntegration": {
    "enabled": true,
    "tagCommits": true,
    "addConvIdToMessage": true
  },
  "search": {
    "caseSensitive": false,
    "maxResults": 50
  }
}

SUMMARY.md Format

# Conversation {conv-id} - TLDR

**Title:** [Brief conversation goal, set via /baton rename, max 60 chars]
**Status:** Active | Completed | Paused
**Started:** YYYY-MM-DD HH:MM
**Duration:** Xh
**Compactions:** N

## Context in 3 Lines
[High-level overview of what's happening]

## Task Checklist
- [x] Completed task
- [ ] Pending task

## Decisions Made
- Decision #N: What was decided and why

## Key Files Created/Modified
- path/to/file.ts (what changed)

## Failed Attempts (Don't Retry)
- Approach X: Why it failed

## Next Actions
1. First priority
2. Second priority

## State Snapshot
**Current file:** exact/path.ts
**Current line:** 42
**Current task:** Specific thing being worked on
**Blockers:** Any blockers
**Ready to:** Next immediate action

Conversation ID Tagging

When adding to shared files (BUGS.md, DECISIONS.md):

**Conv:** conv-20251223-225929

This enables:

• Multiple conversations working simultaneously
• Each conversation identifying their work
• Shared awareness across conversations
• No file conflicts

Integration with CLAUDE.md

Add this to your project's CLAUDE.md:

## Context Management Protocol

This project uses `.claude/` context management system.

**After compaction:** Automatically reads TLDR summaries
**Manual control:** Use `/baton` skill commands
**Token efficiency:** 25-100x compression (50K→1K tokens)
**Auto-save:** Enabled at 70%, 85%, 95% thresholds

See full protocol in CLAUDE.md for details.

Portability

To use across multiple projects:

1. Option A - Local per project:

   cp -r .claude-code/skills/baton /path/to/other/project/.claude-code/skills/
   

2. Option B - Global symlink:

   # Move to central location
   mv .claude-code/skills/baton /mnt/foundry_project/Claude_skills/
   
   # Symlink from ~/.claude/skills/
   ln -s /mnt/foundry_project/Claude_skills/baton ~/.claude/skills/baton
   

3. Option C - Copy to central location:

   cp -r .claude-code/skills/baton /mnt/foundry_project/Claude_skills/
   ln -s /mnt/foundry_project/Claude_skills/baton ~/.claude/skills/baton
   

Token Efficiency

• Full conversation log: 50,000+ tokens
• TLDR summary: 500-2,000 tokens
• Compression ratio: 25-100x
• Post-compaction read: ~1,000-1,500 tokens total
• Auto-save overhead: ~50 tokens per save
• Enables: Long autonomous sessions without context loss

Benefits

✅ For humans:

• Quick scan of what happened
• Thread view of conversations
• Easy navigation
• Instant search across all work
• Weekly/monthly reports for standups

✅ For Claude:

• Efficient context restoration
• No retry of failed approaches
• Preserves technical details
• Enables multi-conversation work
• Auto-saves prevent context loss

✅ For projects:

• Portable across projects
• Standardized context management
• Reduced token costs (98%+ reduction)
• Better long-running session support
• Git integration for traceability

Version

Baton v2.0 - Enhanced with 11 new features (2025-12-25)

New in v2.0:

1. Auto-save triggers (prevent context loss)
2. Conversation switching (multi-conversation workflow)
3. Smart SUMMARY.md updates (reduce manual overhead)
4. Search across conversations (find past context)
5. Title auto-suggestions (AI-powered naming)
6. Validation & health check (data integrity)
7. Export & reporting (status summaries)
8. Conversation metrics (prove system value)
9. Smart context loading (relevance-based)
10. Git integration (branch association, commit messages)
11. Template customization (project-specific needs)