---

name: journaler description: Nightly automated memory consolidation that reviews daily chat history, synthesizes key learnings/decisions, updates MEMORY.md, and manages workspace cleanup. Run automatically via cron to maintain long-term context without manual intervention.

Journaler Agent

Nightly automated consolidation of daily activities into long-term memory.

Purpose

The Journaler Agent runs automatically (via cron) to:

1. Review today's chat history - Synthesize key learnings, decisions, and patterns
2. Update MEMORY.md - Add meaningful entries while avoiding redundancy
3. Clean workspace - Remove temporary files and completed project artifacts
4. Archive projects - Move completed work to appropriate archive folders

This ensures MEMORY.md stays current, valuable insights aren't lost, and the workspace remains organized.

When to Use

• Automatic nightly execution via cron job (primary use case)
• Manual execution when:
  • Significant milestone reached and you want to capture it now
  • Testing journaler behavior before automation
  • After long absence to catch up on consolidation

Core Workflow

0. Run Enhanced Journaler (Primary Method)

Use the unified journaler script:

# Full run (status update + staleness check + checkpoint updates)
python3 /root/.openclaw/workspace/scripts/journaler.py

# Quick status update only
python3 /root/.openclaw/workspace/scripts/journaler.py --quick

# Preview changes
python3 /root/.openclaw/workspace/scripts/journaler.py --dry-run

This automatically:

• Updates PROJECT_STATE.md with current system status
• Detects stale entries (> 30 days old)
• Updates project checkpoints
• Generates staleness report

0.5. Update PROJECT_STATE.md (Alternative)

If you only need to update PROJECT_STATE.md:

python3 /root/.openclaw/workspace/scripts/update_project_state.py

1. Review Daily Chat History

Access logs from:

• Current session: Use sessions_history tool with today's date range
• All sessions: List recent sessions, filter by date, review each

What to look for:

• New decisions - Tech choices, project directions, preferences
• Learnings - Problems solved, discoveries made
• Patterns - Repeated requests, workflows, preferences
• Context updates - Changes in project status, personal circumstances
• Action items - Tasks created, completed, or deferred

What to ignore:

• Routine operational commands (file reads, simple edits)
• Brief confirmations ("OK", "thanks", etc.)
• Duplicate information already in MEMORY.md
• Transient troubleshooting that didn't lead to learnings

2. Synthesize Key Learnings

For each learning point, create a concise entry:

Format:

## YYYY-MM-DD
- **[Category]**: [Brief description]
  - **Details**: [What happened, why it matters]
  - **Action Taken**: [What was done or decided]
  - **Related**: [Cross-reference to related entries]

Categories to use:

• Decisions - Technical or project decisions made
• Learnings - Discoveries, solutions found
• Preferences - Jeff's preferences, patterns
• Updates - Status changes, new information
• Tools - New tools, configurations, workflows

Example entries:

## 2026-02-11
- **Decision**: Use session-based separation for OpenClaw Discord agents
  - **Details**: Research showed no public discussion of multi-agent Discord setups. OpenClaw's session system likely provides separate memory contexts.
  - **Action Taken**: Will configure Discord channels/threads to map to sessions for agent separation
  - **Related**: OpenClaw Discord integrations

- **Learning**: Letta not needed yet for memory enhancement
  - **Details**: Current MEMORY.md + research archive works well. Letta adds complexity for problems Jeff doesn't have yet (semantic search, auto-consolidation).
  - **Action Taken**: Will build simple nightly journaler + semantic search instead (80% value, 10% complexity)
  - **Related**: Memory architecture decisions

3. Update MEMORY.md

Before writing:

1. Read current MEMORY.md to understand existing structure
2. Check for duplicate entries (avoid adding redundant info)
3. Maintain chronological order (newest entries at top)

Writing approach:

• Add new entries under ## YYYY-MM-DD section
• Keep entries concise (2-3 lines of details max)
• Cross-reference related entries when relevant
• Remove outdated entries if superseded by new information

IMPORTANT: MEMORY.md is for decisions and preferences, not project status.

• Project status → PROJECT_STATE.md
• Decisions/preferences → MEMORY.md
• Raw daily logs → memory/YYYY-MM-DD.md

Example section in MEMORY.md:

# MEMORY.md - Long-Term Context

## 2026-02-12
- **Decision**: Implemented nightly journaler for automatic memory consolidation
  - **Details**: Journaler agent runs via cron to synthesize daily learnings, update MEMORY.md, and clean workspace.
  - **Action Taken**: Created journaler skill + semantic search script
  - **Related**: Memory architecture

## 2026-02-11
- [Previous entries...]

4. Clean Workspace

Identify temp files to remove:

• *.tmp, *.temp, *.bak files older than 1 day
• Debug logs in /tmp/ older than 3 days
• Test outputs not needed after review

Safe cleanup only:

• Confirm files are not critical before deletion
• Archive (move to archive/) instead of delete for important outputs
• Keep files less than 1 day old (might still be in use)

Cleanup script: See scripts/cleanup_workspace.py

5. Archive Completed Projects

Identify completed projects:

• Research reports that are no longer actively used
• Old research_report_*.md files that have been consolidated
• Temporary testing directories

Archive approach:

• Move to /root/.openclaw/workspace/archive/research/YYYY-MM-DD/
• Keep INDEX.md updated with archive locations
• Add cross-reference in MEMORY.md if project was significant

Script Reference

scripts/cleanup_workspace.py

Cleans temporary files and archives completed projects.

Usage:

python3 scripts/cleanup_workspace.py --dry-run
python3 scripts/cleanup_workspace.py

What it does:

• Scans workspace for temp files older than 1 day
• Moves completed research reports to archive
• Reports what will be cleaned (use --dry-run to preview)
• Creates archive directory structure if missing

Safety:

• Never deletes files in skills/, research/, or memory/ directories
• Always moves to archive instead of deleting (except /tmp/)
• Logs all actions for audit trail

Manual Execution

To run journaler manually:

# Review last 24 hours of activity
# Synthesize learnings and update MEMORY.md
# Clean workspace and archive projects

Just describe what you want reviewed (e.g., "Synthesize today's learnings") and the journaler will:

1. Review recent chat history
2. Identify key learnings/decisions
3. Update MEMORY.md appropriately
4. Clean workspace

Cron Configuration

Automatic nightly execution:

# Run journaler daily at 2 AM Eastern (7 AM UTC)
0 7 * * * /root/.openclaw/workspace/skills/journaler/scripts/run_journaler.sh >> /var/log/journaler.log 2>&1

Cron script (scripts/run_journaler.sh):

• Sets up environment (Python path, etc.)
• Invokes journaler via CLI or Python
• Logs output to /var/log/journaler.log
• Handles errors gracefully (notifies if consolidation fails)

Quality Guidelines

What Makes a Good Journal Entry

Do include:

• Decision points - Why X was chosen over Y
• Problem → Solution - What broke and how it was fixed
• Pattern recognition - "Jeff prefers X over Y"
• Context updates - "Project now uses [new technology]"
• Cross-references - Links to related topics

Don't include:

• Routine operations that didn't yield insights
• Duplicate information already present
• Trivial details ("created a file", "ran a command")
• Transient debugging that was abandoned

Update Frequency

• High-velocity days (many decisions/learnings): Add detailed entries
• Low-velocity days (routine work): Brief summary or skip if nothing significant
• Milestone reached (project completion): Add consolidation entry summarizing key outcomes

Memory Decay

Review older entries periodically (journaler does this automatically):

• Mark outdated info as superseded
• Consolidate fragmented notes into coherent summaries
• Remove entries no longer relevant to current context

Troubleshooting

If journaler fails:

1. Check logs: /var/log/journaler.log
2. Verify session access: Ensure sessions_history can retrieve logs
3. Test manual run: Run journaler manually to see error output
4. Check MEMORY.md permissions: Ensure writeable

If MEMORY.md gets cluttered:

1. Run journaler manually with --consolidate flag
2. Journaler will merge duplicate entries
3. Remove outdated information
4. Clean up cross-reference structure