Session Launcher
Purpose
Zero context loss between conversations. When a new session starts, this skill:
- Searches for project context files
- Loads memory indexes and past decisions
- Reads session state
- Creates visual dashboard with actionable links
- Makes you immediately productive
For SDAM users: Compensates for lack of episodic memory by externalizing all session context. For ADHD users: Eliminates "where was I?" friction - instant continuation. For dyschronometria: Shows explicit timestamps on all context.
Activation Triggers
- User says: "hi-ai"
- User says: "continue", "restore context", "what was I working on"
- Start of new conversation (proactively offer)
Core Workflow
1. Search for Context Files
Check these locations in order:
# Current project context
./.context/current.json
./.context/session.json
./DECISIONS.md
./README.md
# User's memory storage
~/.claude-memories/ (Linux/macOS) or %USERPROFILE%\.claude-memories\ (Windows)index.json
~/.claude-sessions/ (Linux/macOS) or %USERPROFILE%\.claude-sessions\ (Windows)current.json
~/.claude-sessions/ (Linux/macOS) or %USERPROFILE%\.claude-sessions\ (Windows)projects/2. Load Memory Index
Read ~/.claude-memories/ (Linux/macOS) or %USERPROFILE%\.claude-memories\ (Windows)index.json:
{
"total_memories": N,
"memories": [
{
"id": "unique-id",
"type": "DECISION|BLOCKER|CONTEXT|PREFERENCE|PROCEDURE",
"content": "the memory",
"timestamp": "ISO8601",
"tags": ["tag1", "tag2"],
"project": "project-name"
}
]
}Extract:
- Recent decisions (last 7 days)
- Active blockers (status!= "resolved")
- Project context (matching current directory)
- User preferences
3. Identify Current Project
Determine project from:
- Current working directory name
.context/project.txtif exists- Git repository name:
basename $(git rev-parse --show-toplevel 2>/dev/null) - Ask user if unclear
4. Create Session Dashboard
Generate HTML file at ~/.claude-artifacts/ (Linux/macOS) or %USERPROFILE%\.claude-artifacts\ (Windows)session-dashboard-{timestamp}.html:
<!DOCTYPE html>
<html>
<head>
<meta charset="UTF-8">
<title>Session Restored</title>
<style>
body {
font-family: monospace;
max-width: 1200px;
margin: 40px auto;
padding: 20px;
background: #1e1e1e;
color: #d4d4d4;
}
.section {
background: #252526;
border: 1px solid #3e3e42;
border-radius: 6px;
padding: 20px;
margin: 20px 0;
}
.section h2 {
margin-top: 0;
color: #4ec9b0;
border-bottom: 1px solid #3e3e42;
padding-bottom: 10px;
}
.timestamp {
color: #858585;
font-size: 0.9em;
}
.decision {
background: #2d2d30;
padding: 10px;
margin: 10px 0;
border-left: 3px solid #4ec9b0;
}
.blocker {
background: #2d2d30;
padding: 10px;
margin: 10px 0;
border-left: 3px solid #f48771;
}
.action {
display: inline-block;
background: #0e639c;
color: white;
padding: 8px 16px;
margin: 5px;
border-radius: 4px;
text-decoration: none;
}
.metric {
font-size: 2em;
color: #4ec9b0;
font-weight: bold;
}
</style>
</head>
<body>
<h1>🚀 Session Restored</h1>
<p class="timestamp">Restored at: {current_timestamp}</p>
<div class="section">
<h2>📁 Current Project</h2>
<div class="metric">{project_name}</div>
<p>Location: <code>{project_path}</code></p>
<p>Last worked: <span class="timestamp">{last_session_time}</span></p>
</div>
<div class="section">
<h2>🎯 Recent Decisions (Last 7 Days)</h2>
{for each recent decision:}
<div class="decision">
<strong>{decision.content}</strong>
<div class="timestamp">{decision.timestamp} - {relative_time}</div>
{if decision.tags:}<p>Tags: {tags_joined}</p>{endif}
</div>
{endfor}
</div>
<div class="section">
<h2>🚧 Active Blockers</h2>
{for each active blocker:}
<div class="blocker">
<strong>{blocker.content}</strong>
<div class="timestamp">{blocker.timestamp} - {relative_time}</div>
</div>
{endfor}
{if no blockers:}<p>✅ No active blockers</p>{endif}
</div>
<div class="section">
<h2>📊 Memory Stats</h2>
<p>Total memories: <span class="metric">{total_memories}</span></p>
<p>Decisions: {decision_count} | Blockers: {blocker_count} | Procedures: {procedure_count}</p>
</div>
<div class="section">
<h2>⚡ Quick Actions</h2>
<a href="file://~/.claude-memories/ (Linux/macOS) or %USERPROFILE%\.claude-memories\ (Windows)index.json" class="action">View All Memories</a>
<a href="file://{project_path}" class="action">Open Project</a>
<a href="file://~/.claude-artifacts/ (Linux/macOS) or %USERPROFILE%\.claude-artifacts\ (Windows)" class="action">View Artifacts</a>
</div>
</body>
</html>5. Output Summary
After creating dashboard, output to user:
🚀 Session Restored
📁 Project: {project_name}
📍 Location: {project_path}
🕐 Last worked: {relative_time}
📝 Recent Decisions ({count}):
{for each decision (max 5):}
• {decision.content} ({relative_time})
{endfor}
🚧 Active Blockers ({count}):
{for each blocker:}
• {blocker.content} ({relative_time})
{endfor}
💾 Memory: {total_memories} total memories loaded
📊 Dashboard: {dashboard_path}
✅ Full context restored. Ready to continue!
What would you like to work on?Memory Integration
Reading Memories
Always check memory index for:
- Project-specific context
- User preferences (tech stack, coding style)
- Past decisions affecting current work
- Similar past problems/solutions
Time Anchoring (for Dyschronometria)
Convert all timestamps to relative format:
- "2 hours ago"
- "3 days ago"
- "Last Tuesday"
- Include ISO timestamp for precision
Context Filtering
Filter memories by:
- Project match:
memory.project === current_project - Recency: Last 30 days weighted higher
- Type priority: BLOCKER > DECISION > CONTEXT > PROCEDURE
- Relevance: Tag matching if available
Session State Management
Saving Session State
Before ending conversation (if user says "goodbye", "see you", "done for now"):
- Create/update
~/.claude-sessions/ (Linux/macOS) or %USERPROFILE%\.claude-sessions\ (Windows)current.json:
{
"project": "project-name",
"project_path": "/full/path",
"last_active": "ISO8601_timestamp",
"last_files": [
"/path/to/file1.js",
"/path/to/file2.py"
],
"last_topic": "Brief description of what we were working on",
"next_actions": [
"Suggested next step 1",
"Suggested next step 2"
]
}- Update memory index with session summary:
{
"type": "NOTE",
"content": "Session ended: {brief_summary_of_work}",
"timestamp": "ISO8601",
"tags": ["session", "project-name"],
"project": "project-name"
}Project-Specific Sessions
For multi-project users, maintain:
~/.claude-sessions/ (Linux/macOS) or %USERPROFILE%\.claude-sessions\ (Windows)projects/
├── boostbox.json
├── toolhub.json
├── 88dimensions.json
└── mcp-stack.jsonEach contains project-specific state.
Visual Dashboard Features
For Aphantasia Users
Always provide:
- Concrete metrics (numbers, counts)
- Visual structure (HTML dashboard)
- Explicit links (clickable file paths)
- Status indicators (✅ ❌ 🚧 emojis)
For ADHD Users
Dashboard must:
- Load instantly (<2 seconds)
- Show quick actions (immediate next steps)
- Highlight blockers (red/urgent visual treatment)
- Provide direct links (one-click access to files)
For SDAM Users
Dashboard must:
- Externalize all context (no assumption of memory)
- Show time anchors (when things happened)
- Display decision history (what was decided)
- Link to artifacts (proof of past work)
Error Handling
If No Context Found
👋 Hi! This looks like a fresh start.
I couldn't find existing session context, so let's establish some:
1. What project are you working on?
2. Should I create a memory index for you?
3. Would you like to start tracking decisions?
(I'll create the necessary directories and get you set up)If Memory Index Corrupted
⚠️ Memory index appears corrupted.
Creating backup at: ~/.claude-memories/ (Linux/macOS) or %USERPROFILE%\.claude-memories\ (Windows)backups/index-{timestamp}.json
Initializing fresh index...
✅ Fresh index created. Previous memories backed up.If Multiple Projects Detected
🤔 I found context for multiple projects:
1. BOOSTBOX (last active: 2 days ago)
2. Tool Hub (last active: 5 days ago)
3. MCP Stack (last active: 1 week ago)
Which project would you like to work on?Integration with Other Skills
Memory Manager
- Queries memory-manager for structured memory search
- Defers complex memory operations to memory-manager
- Uses memory-manager's categorization system
Context Manager
- Works with context-manager to maintain
.context/directory - Reads context-manager's structured data
- Updates session state via context-manager
Error Debugger
- Surfaces recent blockers for error-debugger to consider
- Checks if current error matches past error memories
Quick Reference
File Locations
- Memory index:
~/.claude-memories/ (Linux/macOS) or %USERPROFILE%\.claude-memories\ (Windows)index.json - Current session:
~/.claude-sessions/ (Linux/macOS) or %USERPROFILE%\.claude-sessions\ (Windows)current.json - Project sessions:
~/.claude-sessions/ (Linux/macOS) or %USERPROFILE%\.claude-sessions\ (Windows)projects/{name}.json - Dashboards:
~/.claude-artifacts/ (Linux/macOS) or %USERPROFILE%\.claude-artifacts\ (Windows)session-dashboard-*.html
Trigger Phrases
- "hi-ai" (primary)
- "continue"
- "restore context"
- "what was I working on"
- "where did we leave off"
Success Criteria
✅ User says "hi-ai" and is immediately productive ✅ No need to explain "what we were working on" ✅ All past decisions visible ✅ Active blockers surfaced ✅ Context restored in <2 seconds