Self-Improving Agent
Description
Self-improving agent system that learns from its own errors. Automatically detects issues, researches solutions, and implements improvements.
Based on production-ready patterns: Reflection, PEV (Plan-Execute-Verify), and Meta-Controller architectures.
Features
- Auto-Trigger: Automatically runs when new errors are detected
- Topic Selection: Analyzes error patterns and selects high-priority topics
- Impact Measurement: Records before/after metrics to measure improvement effectiveness
- Procedural Memory: Remembers working commands/scripts between sessions
What It Does
- Error Detection: Monitors cron jobs, circuit breakers, and error logs
- Topic Selection: Chooses research topics based on error impact
- Research: Generates research files with improvement recommendations
- Backlog: Creates high-impact tasks in backlog
- Execution: Backlog Agent PM executes tasks sorted by impact
- Measurement: Records before/after to measure effectiveness
- Learning: Remembers what worked for future reference
Architecture
Errors → Auto-Trigger → Topic Selector → Research → Backlog → Agent PM
↓ ↓
Impact Measurement ← Procedural Memory
Usage
Run Full Cycle
python3 skills/self-improving-agent/scripts/self_improvement_cycle.py
Check Errors Only
python3 skills/self-improving-agent/scripts/topic_selector.py
Record Impact
python3 skills/self-improving-agent/scripts/impact_measurement.py \
--record \
--task "Fix cron timeout" \
--before '{"error_count": 5}' \
--after '{"error_count": 0}'
Search Procedural Memory
python3 skills/self-improving-agent/scripts/procedural_memory.py --search "backup"
Cron Integration
Add to your cron jobs:
{
"name": "Self-Improvement",
"schedule": "0 10 * * *",
"command": "python3 skills/self-improving-agent/scripts/self_improvement_cycle.py"
}
Requirements
- Python 3.10+
- OpenClaw workspace at /root/.openclaw/workspace
- Write access to memory/ directory
Files
scripts/auto_trigger.py- Auto-triggers on new errorsscripts/impact_measurement.py- Measures improvement impactscripts/procedural_memory.py- Stores working proceduresscripts/self_improvement_cycle.py- Main orchestratorscripts/topic_selector.py- Selects topics by priority
Created By
- Based on production patterns: Reflection, PEV, Meta-Controller
- Tested in production environment
- Ready for ClawHub publication