project-context-sync

# Project Context Synchronization

This skill ensures Claude maintains accurate project state across sessions, following patterns from Anthropic's long-running agent research.

## Core Philosophy

Each session should:
1. **Start informed**: Read existing context before diving in
2. **Work incrementally**: One feature at a time, commit often
3. **End verified**: Run verification commands before marking work complete
4. **End clean**: Leave documentation that your future self (next session) can understand

## Context Files

### .claude/PROGRESS.md

This is the **semantic bridge** between sessions. It answers: "What was I doing? What's next?"

Required sections:
- **Current State**: What's working, what's broken
- **Recent Work**: Last 3-5 significant changes with rationale
- **Next Steps**: Prioritized list of what to work on
- **Blockers** (optional): Known issues or decisions needed
- **Architecture Decisions** (optional): Why significant choices were made

### .claude/feature_list.json (For Multi-Session Features)

When working on features that span multiple sessions, use this structured format:

```json
{
  "feature": "User Authentication",
  "description": "Add JWT-based authentication with login/logout",
  "created": "2025-01-15T10:00:00Z",
  "status": "in-progress",
  "estimatedSessions": 4,
  "completedSessions": 1,
  "items": [
    {
      "id": "item-001",
      "description": "Create JWT token generation utility",
      "status": "complete",
      "acceptanceCriteria": [
        "Generates valid JWT tokens",
        "Tokens include user ID and expiration"
      ],
      "verification": [
        {
          "command": "npm test -- --grep 'JWT'",
          "description": "JWT tests pass"
        }
      ],
      "completedAt": "2025-01-15T14:30:00Z",
      "sessionId": "abc123"
    },
    {
      "id": "item-002",
      "description": "Create login API endpoint",
      "status": "in-progress",
      "acceptanceCriteria": [
        "POST /api/login accepts em