How do I install the Lev Self skill?

Run npx skillmds@latest add lev-os/lev-self in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Lev Self safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Lev Self?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Lev Self free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Lev Self?

lev-os (@lev-os) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Lev Self by lev-os — Agent Skill

Name: Lev Self
Rating: 4.666666666666667 (3 reviews)
Author: lev-os

Lev Self

[WHAT] Self-improvement system transforming event streams into actionable system improvements [HOW] Analyze events.jsonl for patterns, generate proposals with confidence scores, human-approve, apply to skills/config/daemons [WHEN] Use when asking what lev can learn, proposing improvements, or conducting fail-forward root cause analysis [WHY] Enables proactive system evolution based on observed failures and patterns instead of reactive fixes Triggers: "what can lev learn", "propose improvements", "self-learn", "why did X fail", "improve lev", "fail-forward"

lev-os c1f5538 5 files · 9.9 KB Updated Sep 10, 2026

File contents

Lev Self-Improvement

Transform event streams into actionable system improvements. Proactive learning (not just reactive fixes).

Quick Reference

Trigger	Action
"what can lev learn?"	Full analysis cycle
"propose improvements"	Generate proposals from patterns
"why did X fail?"	Root cause analysis (fail-forward)
"self-learn"	Automated improvement cycle

Architecture

events.jsonl → Analyze → Patterns → Proposals → Human Review → Apply
                 ↑                                              │
                 └──────────── feedback loop ──────────────────┘

Core Workflows

1. Event Analysis

lev learn analyze              # Detect patterns in events.jsonl
lev learn analyze --since 24h  # Last 24 hours only

See: references/event-analysis.md

2. Fail-Forward Protocol

When something fails, extract learning:

Capture: What happened? (exact error, context)
Root Cause: Why? (5 whys, dependencies)
Proposal: How to prevent? (skill patch, config change)
Confidence: How sure? (low/medium/high)

See: references/fail-forward.md

3. Proposal Workflow

# ~/.config/lev/proposals/{id}.yaml
id: prop-abc123
type: skill-patch | config-change | new-workflow
target: ~/.claude/skills/lev/SKILL.md
confidence: 0.85
description: "Add timeout handling"
diff: |
  + timeout: 30s
status: pending | approved | rejected | applied

See: references/proposals.md

Evolution Targets

Target	Location	When
Skills	~/.claude/skills/*/SKILL.md	Behavior improvements
Config	~/lev/.lev/config.yaml	Settings optimization
Daemons	~/.config/lev/daemons.yaml	Process tuning
Workflows	~/lev/workflows/*.yaml	Pattern codification

Relationship to Other Skills

skill-evolver (absorbed): Reactive skill fixes
lev (sibling): Behavior definition (lev-self improves it)
bd (integration): Track proposals as issues

Storage Schema

~/.config/lev/
├── proposals/           # Pending proposals
│   └── {id}.yaml
├── memory/
│   ├── patterns.jsonl   # Detected patterns
│   ├── proposals.jsonl  # Proposal history
│   └── applied.jsonl    # Applied changes + outcomes
└── patterns.jsonl       # Active patterns for matching

Confidence Thresholds

Confidence	Action
≥90%	Auto-apply (after notification)
70-89%	Propose with recommendation
50-69%	Propose, request review
<50%	Log only, don't propose

See: references/tracking-schema.md

CLI Quick Reference

# Analysis
lev learn analyze             # Full event analysis
lev learn patterns            # Show detected patterns

# Proposals
lev learn propose             # Generate proposals from patterns
lev learn review              # Interactive proposal review
lev learn apply <id>          # Apply approved proposal

# Tracking
lev learn status              # Show pending proposals
lev learn history             # Show applied changes

Human-in-the-Loop

All changes require confirmation:

Proposal generated → notification
Human reviews diff
Approve/reject/modify
Applied changes logged with outcome

Rollback:

lev learn rollback <id>       # Revert applied proposal

For detailed schemas and protocols, see references/

---name: lev-selfdescription: [WHAT] Self-improvement system transforming event streams into actionable system improvements [HOW] Analyze events.jsonl for patterns, generate proposals with confidence scores, human-approve, apply to skills/config/daemons [WHEN] Use when asking what lev can learn, proposing improvements, or conducting fail-forward root cause analysis [WHY] Enables proactive system evolution based on observed failures and patterns instead of reactive fixes Triggers: "what can lev learn", "propose improvements", "self-learn", "why did X fail", "improve lev", "fail-forward"---# Lev Self-ImprovementTransform event streams into actionable system improvements. Proactive learning (not just reactive fixes).## Quick Reference| Trigger | Action ||---------|--------|| "what can lev learn?" | Full analysis cycle || "propose improvements" | Generate proposals from patterns || "why did X fail?" | Root cause analysis (fail-forward) || "self-learn" | Automated improvement cycle |## Architecture```events.jsonl → Analyze → Patterns → Proposals → Human Review → Apply ↑ │ └──────────── feedback loop ──────────────────┘```## Core Workflows### 1. Event Analysis```bashlev learn analyze # Detect patterns in events.jsonllev learn analyze --since 24h # Last 24 hours only```See: `references/event-analysis.md`### 2. Fail-Forward ProtocolWhen something fails, extract learning:1. Capture: What happened? (exact error, context)2. Root Cause: Why? (5 whys, dependencies)3. Proposal: How to prevent? (skill patch, config change)4. Confidence: How sure? (low/medium/high)See: `references/fail-forward.md`### 3. Proposal Workflow```yaml# ~/.config/lev/proposals/{id}.yamlid: prop-abc123type: skill-patch | config-change | new-workflowtarget: ~/.claude/skills/lev/SKILL.mdconfidence: 0.85description: "Add timeout handling"diff: | + timeout: 30sstatus: pending | approved | rejected | applied```See: `references/proposals.md`## Evolution Targets| Target | Location | When ||--------|----------|------|| Skills | ~/.claude/skills/*/SKILL.md | Behavior improvements || Config | ~/lev/.lev/config.yaml | Settings optimization || Daemons | ~/.config/lev/daemons.yaml | Process tuning || Workflows | ~/lev/workflows/*.yaml | Pattern codification |## Relationship to Other Skills- **skill-evolver** (absorbed): Reactive skill fixes- **lev** (sibling): Behavior definition (lev-self improves it)- **bd** (integration): Track proposals as issues## Storage Schema```~/.config/lev/├── proposals/ # Pending proposals│ └── {id}.yaml├── memory/│ ├── patterns.jsonl # Detected patterns│ ├── proposals.jsonl # Proposal history│ └── applied.jsonl # Applied changes + outcomes└── patterns.jsonl # Active patterns for matching```## Confidence Thresholds| Confidence | Action ||------------|--------|| ≥90% | Auto-apply (after notification) || 70-89% | Propose with recommendation || 50-69% | Propose, request review || <50% | Log only, don't propose |See: `references/tracking-schema.md`## CLI Quick Reference```bash# Analysislev learn analyze # Full event analysislev learn patterns # Show detected patterns# Proposalslev learn propose # Generate proposals from patternslev learn review # Interactive proposal reviewlev learn apply <id> # Apply approved proposal# Trackinglev learn status # Show pending proposalslev learn history # Show applied changes```## Human-in-the-Loop**All changes require confirmation:**1. Proposal generated → notification2. Human reviews diff3. Approve/reject/modify4. Applied changes logged with outcome**Rollback:**```bashlev learn rollback <id> # Revert applied proposal```---*For detailed schemas and protocols, see references/*

Frequently asked questions