Project Context Keeper v3

A Claude skill for managing development context across coding sessions with progressive timeline tracking, auto-activation, and incremental saves.

What's New in v3

Auto-Activation

• No more manual context: start commands for most work
• Automatically detects when you're working on GitHub projects
• Silently loads context and shows condensed status
• "Just start coding" - context management happens in background

Incremental Saves (Checkpoints)

• Save individual features/milestones within a session
• Use context: checkpoint or let Claude detect completion signals
• Each checkpoint captures what was done + files touched
• Session end consolidates all checkpoints into full session entry

Smarter Triggers

• Detects GitHub paths, project names, Git operations
• Recognizes completion signals: "that works", "feature done", etc.
• Offers checkpoints without interrupting flow
• Multi-project awareness (caches active projects)

Quick Start

First Time Setup

1. Start working on a GitHub project
2. Claude will detect and ask: "Should I track context for this project?"
3. Provide project name, goal, tech stack
4. Context file created at [project-root]/devbridge-context.md

Daily Usage

Start working:

User: "I need to add OAuth to anchor-loop"
Claude: 📊 Anchor Loop • in_progress • Last: 2 days ago
        ✅ Last: Set up Supabase auth
        🎯 Next: OAuth configuration
        
        Starting OAuth work...

Mid-session checkpoint:

User: "Google OAuth is working"
Claude: Nice! Checkpoint this? [Yes/No]
User: Yes
Claude: ✅ Checkpoint saved. Session continues.

End session:

User: context: save
Claude: [Interviews for session summary, including all checkpoints]
        ✅ Context saved. Session 42 added to timeline.

Commands

Primary

• Auto-start - Just mention your GitHub project
• context: save - End session, save to timeline
• context: checkpoint - Save progress mid-session
• context: history - View past sessions
• context: insights - Velocity metrics and patterns

Utility

• context: status - Quick state check
• context: skip - Don't load context this time
• context: automate - Generate continuous-claude command

File Structure

Context is stored as devbridge-context.md in your project root:

project-root/
├── devbridge-context.md    ← Your project memory
├── .devbridge/
│   └── automation-notes.md ← Automation logs (if used)
└── [your code]

The context file contains:

• Current State - Latest snapshot (overwritten each save)
• Timeline - Append-only session history with checkpoints
• Insights - Auto-computed velocity and pattern analysis

Schema

Session Entry

• Timestamp, duration, state changes
• Completed work summary
• Checkpoint entries (if any)
• Key decisions with rationale
• Blockers resolved with time spent
• Files touched + Git commits
• Next priority + notes for future

Checkpoint Entry

• Timestamp within session
• What was completed (150 char max)
• Files modified
• Optional decision note

Best Practices

For Checkpoints

• One feature = one checkpoint
• Don't over-checkpoint (3-5 per multi-hour session)
• Capture decisions while fresh
• Update files list accurately

For Context Management

• Let auto-trigger handle loading
• Use explicit context: start only when auto-detection fails
• Commit context file to Git (it's project memory)
• Review context: insights quarterly for patterns

For Timeline

• Never delete entries (append-only)
• Be specific in completions
• Document decision rationale
• Track blocker resolution time

Integration

With continuous-claude

• Generate automation commands via context: automate
• Context is injected into automation prompts
• Automation sessions saved to timeline with cost tracking
• Resume with context: resume-automation

With Git

• Context file is markdown (Git-friendly)
• Commit after significant milestones
• Include in .gitignore if context contains sensitive info
• Checkpoint helps tie context to specific commits

Migration

From v2

• Automatic detection and upgrade
• Backup created: devbridge-context-v2-backup.md
• All data preserved
• New checkpoint structure added
• Future saves support checkpoints

From v1

• Two-step migration (v1 → v2 → v3)
• Snapshot converted to timeline entry
• All existing data preserved

Philosophy

1. Timeline, Not Snapshot - History is append-only
2. Auto-Activated - Context management fades into background
3. Incremental Progress - Checkpoint granular wins
4. Decision Context - Capture WHY, not just WHAT
5. Human-Led - You summarize, AI structures
6. Token-Efficient - Character limits prevent bloat

Examples

See SKILL.md for detailed examples of:

• Auto-start workflows
• Mid-session checkpointing
• Session end with multiple features
• Automation setup and monitoring
• Timeline analysis and insights

License

Internal use only.

Version

Current: v3.0
Compatible with: v2.x, v1.x (with auto-migration)