context-restoration

# Context Restoration Skill

## Purpose

Restore complete session state after context loss events (auto-compact, session break, interruption).

**Zero Context Loss Promise**: When combined with context-checkpoint skill, users can resume exactly where they left off regardless of auto-compact events or session boundaries.

---

## Inputs

**Required:**
- `checkpoint_id` (string): Unique checkpoint identifier (e.g., "SHANNON-W2-20251103T143000")

**Optional:**
- `auto_select` (boolean): Automatically select most recent checkpoint if ID not provided (default: true)
- `project_id` (string): Filter checkpoints by specific project
- `validate_integrity` (boolean): Verify checkpoint integrity via SHA-256 hash (default: true)

---

## When to Use This Skill

Activate context-restoration when:

1. **After Auto-Compact** - Claude Code compressed conversation history
2. **New Session Start** - User returns after hours/days
3. **Context Loss Detection** - Claude doesn't remember recent work
4. **Project Switching** - Loading different project context
5. **Error Recovery** - Restoring from last known good state

---

## Workflow

### Step 1: Identify Checkpoint

```yaml
Input Options:
  - Checkpoint ID: "precompact_checkpoint_20250930T143000Z"
  - Auto Mode: Find most recent checkpoint automatically
  - Project-Specific: Latest checkpoint for specific project
  - Phase-Specific: Checkpoint from specific phase
```

**Auto-Detection Logic**:
```python
# Priority order for auto-selection:
1. Read "latest_checkpoint" pointer from Serena
2. If not found, list all checkpoints and sort by timestamp
3. Select most recent checkpoint
4. Validate checkpoint is not corrupted
```

### Step 2: Retrieve Checkpoint from Serena

**Serena MCP Operation**:
```python
# Method 1: Direct retrieval (if ID known)
checkpoint_data = read_memory(checkpoint_id)

# Method 2: Auto-selection
latest_key = read_memory("latest_checkpoint")
checkpoint_data = read_memory(latest_key)

# Method 3: Search and filter
a