What data does Lev Self-Improvement analyze?

It primarily monitors events.jsonl for performance patterns and stores its learning history, detected patterns, and pending proposals in the ~/.config/lev/ directory.

Can I revert a change made by the self-improvement system?

Yes, the skill includes a dedicated rollback command that allows you to revert any applied proposal if the outcome does not meet expectations.

How does the fail-forward protocol work?

When a task fails, the skill captures the context, performs a root cause analysis, and generates a proposal (such as a skill patch) to prevent that specific error from occurring in the future.

Does this skill make changes to my system automatically?

Only improvements with a confidence score of 90% or higher are auto-applied after notification. Most proposals (50-89% confidence) require human review and approval before they are implemented.

What is the primary goal of Lev Self-Improvement?

It transforms event logs into actionable improvements for your agent's skills and configurations, enabling the system to evolve and optimize itself based on its own performance history.

Lev Self-Improvement

Name: Lev Self-Improvement
Author: lev-os

bylev-os

0•

분석 및 모니터링

Transforms agent event streams into actionable system improvements through proactive pattern analysis and automated evolution.

Lev Self-Improvement is a meta-skill designed to evolve an agent's behavior by analyzing its own performance data. It parses event logs to identify recurring patterns and failures, generating structured improvement proposals for skills, configurations, and daemons. By implementing a 'fail-forward' protocol and a confidence-based application system, it ensures that Claude becomes more efficient and accurate over time, shifting from reactive troubleshooting to proactive system optimization with human-in-the-loop oversight.

주요 기능

01Automated Proposal Generation: Creates structured YAML proposals for skill patches, config changes, and workflow updates.

02Event Stream Analysis: Detects patterns and anomalies in system-generated event logs (events.jsonl).

03Recursive Evolution: Targets the agent's own skills and environment settings for continuous performance tuning.

040 GitHub stars

05Fail-Forward Protocol: Conducts automated root cause analysis (5 Whys) after task failures to prevent recurrence.

06Confidence-Based Governance: Uses scoring thresholds to determine if an improvement should be auto-applied or reviewed by a human.

사용 사례

01Automating the 'post-mortem' process after a CLI command or workflow failure.

02Proposing configuration optimizations based on observed interaction patterns.

03Analyzing historical logs to identify and fix recurring execution bottlenecks.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add lev-os/agents lev-self

For use in Claude.ai and ChatGPT

Download Skill