Context Death & Recovery

Context death happens when AI agents lose their session memory due to crashes, resets, or context window limits. ClawVault's checkpoint/recover/wake cycle ensures continuity across these inevitable interruptions.

What is Context Death?

Context death occurs when:

Session crashes due to API errors or system issues
Manual resets with /new command
Context window overflow forcing conversation pruning
System restarts or process termination
Agent handoffs between different sessions

Without proper handling, agents lose:

What they were working on
Recent decisions made
Current blockers and next steps
Conversational context and progress

The Recovery Cycle

ClawVault provides a three-part solution:

1. Checkpoint (Before Death)

Save current state during work:

clawvault checkpoint --working-on "fixing auth bug" --focus "token validation" --blocked "waiting for API docs"

What's saved:

Current task summary
Key focus areas
Known blockers
Session metadata (timestamp, key)

2. Recover (After Death)

Check if previous session died unexpectedly:

clawvault recover --clear

Recovery info includes:

Death detection (was previous session interrupted?)
Last checkpoint details
Time since last activity
Handoff recommendations

3. Wake (New Session Start)

Comprehensive context restoration:

clawvault wake

Wake combines:

Recovery check + flag clearing
Recent handoff recap
Critical observations summary
Executive summary (LLM-generated)

Dirty Death Flag

ClawVault tracks unexpected session termination with a .clawvault/dirty-death.flag file.

Flag created when:

Session ends without proper sleep handoff
Process terminated abruptly
/new command used without checkpoint

Flag cleared when:

clawvault wake or recover --clear runs
Proper sleep handoff completed

OpenClaw Hook Auto-Detection

The ClawVault hook automatically handles context death:

gateway:startup Event

// Detects dirty death on OpenClaw startup
if (await hasDirtyDeath(vaultPath)) {
 const recovery = await recover(vaultPath);
 injectContextDeathAlert(recovery);
}

Alert injection:

 **Context Death Detected** 

Previous session ended unexpectedly at 14:32.
Last checkpoint: "fixing auth bug" (focus: token validation)
Blockers: waiting for API docs

Run `clawvault wake` for full context recovery.

command:new Event

// Auto-checkpoint before /new reset
await checkpoint({
 workingOn: inferCurrentWork(),
 sessionKey: getSessionId()
});

This ensures context is saved even if you forget to manually checkpoint.

Best Practices

Frequent Checkpointing

Checkpoint every 10-15 minutes during heavy work:

clawvault checkpoint --working-on "PR review" --focus "type safety refactor"

Rich Session Ends

Use sleep for detailed handoffs:

clawvault sleep "completed auth refactor" \
 --next "write integration tests, deploy to staging" \
 --blocked "waiting for code review" \
 --decisions "using JWT over sessions for scalability" \
 --feeling "productive, code is cleaner"

Morning Wake Ritual

Start each session with wake:

clawvault wake

This becomes your daily briefing with:

What happened since last session
Current priorities and blockers
Recent critical decisions
Overall context summary

Manual Recovery

When automatic detection fails, use manual tools:

Check for Death

clawvault recover
# Shows: death detection, last checkpoint, time since activity

Force Clear Flag

clawvault recover --clear
# Clears dirty death flag, shows recovery info

Detailed Recap

clawvault recap --brief
# Recent handoffs, active projects, pending commitments

Recovery Information

A typical recovery report:

## Recovery Status

**Context Death:** Yes (detected)
**Death Time:** 2026-02-13 14:32 (2 hours ago)
**Last Checkpoint:** 14:30 
 - Working on: fixing auth bug
 - Focus: token validation logic 
 - Blocked: waiting for API documentation

**Recent Handoff:** 12:45
 - Summary: authentication refactor 
 - Next: write integration tests
 - Status: 80% complete

**Recommendations:**
- Continue with auth bug fix
- Check if API docs arrived 
- Review recent authentication decisions

Hook Configuration

Enable automatic context death handling:

# Install hook
openclaw hooks install clawvault

# Enable hook 
openclaw hooks enable clawvault

# Verify hook status
openclaw hooks status clawvault

The hook runs on gateway:startup and command:new events to provide seamless protection.

Environment Variables

Control recovery behavior:

# Disable death detection
export CLAWVAULT_NO_DEATH_DETECTION=1

# Custom checkpoint frequency (minutes) 
export CLAWVAULT_CHECKPOINT_INTERVAL=10

# Verbose recovery logging
export CLAWVAULT_VERBOSE_RECOVERY=1

Example Session Flow

# Start work
clawvault wake
# → Shows: previous session died at 14:32, last working on "auth bug"

# Work for a while, then checkpoint 
clawvault checkpoint --working-on "fixing token validation"

# Continue work... then session crashes

# Next session
clawvault wake 
# → Recovery alert: "Previous session died, was fixing token validation"

# Continue where you left off

Session Transcript Repair

When context death causes corrupted session files:

# Check for transcript corruption
clawvault repair-session --dry-run

# Fix corrupted tool_use/tool_result chains
clawvault repair-session

# Repair specific session
clawvault repair-session --session <id> --agent <agent-id>

This fixes common API rejection issues that prevent session continuation.

Context death is inevitable in long-running AI work. Embrace the checkpoint/recover/wake cycle as a daily ritual rather than trying to avoid resets entirely.

The key insight: make context death survivable rather than preventing it. With proper checkpointing and recovery, session resets become refreshing breaks rather than productivity killers.

Context Death & Recovery

On this page