Self-Improve

A pluggable self-improvement framework for AI agents. Automatically learns from mistakes, corrections, and feedback to continuously improve execution quality...

Security Scan

The skill's code and instructions largely match its self-improvement purpose, but it is capable of scanning and writing shared/system files (and adding cron entries) based on user-supplied paths — that combination is powerful and potentially dangerous if misconfigured.

Purpose & Capability

The name/description match the included scripts: they scan feedback, distill rules, maintain HOT/WARM/COLD memories, propose changes, and can create/maintain knowledge files. That capability set is coherent with a self-improvement framework. However the skill is explicitly designed to operate at a system/shared scope (engine text: scanning '/path/to/self-improve' and 'read each agent's session logs (if accessible)') and to write system files (AGENTS.md, openclaw.json, SKILL.md) when proposals are 'solidified'. Those system-level operations are more powerful than a purely local/self-contained improvement tool and warrant caution.

Instruction Scope

SKILL.md and ENGINE.md instruct the agent to run setup, register a cron entry, scan memory logs, and propose or write to system files. The scripts will read and write many files under the configured SELF_IMPROVE_ROOT (feedback, corrections, hot/warm/cold memory, proposals). The framework also states it can create new directories/files without asking and that only 'solidification' requires confirmation — but many non-system outputs are written automatically. If the configured storage paths point to shared/global agent directories (e.g., .openclaw, /path/to/learned, or root of agent system), the skill will read other agents' logs and write to system files, which goes beyond a narrowly scoped tool.

Install Mechanism

No install spec — instruction-only. That reduces supply-chain risk, but the package includes runnable Node.js scripts (setup.mjs, run scripts). Running these local scripts will execute code on disk (no external downloads), so risk shifts to runtime file operations rather than package download vectors.

Credentials

The skill requests no environment variables or external credentials (good). But it relies entirely on user-configured paths (user-config.yaml / storage.root / knowledge_root / workspace_root). Those path settings control what the skill reads/writes; if mis-set to global/system locations, the skill gains broad access. The number of required env/creds is proportionate, but the path configuration gives it broad potential reach.

Persistence & Privilege

always:false (normal), and the skill is user-invocable and can be autonomously invoked by the agent. More importantly, SKILL.md/ENGINE.md state the setup writes a Cron entry (and mentions writing a prompt to Cron configuration), and the framework can create directories/files and update indexes automatically. Combined, these grant the skill persistent, periodic execution and the ability to modify system/shared files — a significant privilege if storage paths are global.

What to consider before installing

This skill contains many local Node scripts that will read and write files under whatever storage root you configure — including the ability to create directories, add cron tasks, and propose changes that can be applied to system files (AGENTS.md, SKILL.md, openclaw.json, etc.). Before installing or running it: 1) Inspect setup.mjs and any code that modifies cron or writes to openclaw.json to understand exactly what will be changed. 2) Ensure user-config.yaml storage.root and knowledge_root point to an isolated directory you control (not the global .openclaw, agent home, or system config directories). 3) Run the skill in a sandbox or test environment first. 4) Confirm that the approval workflow actually prevents automatic writes to critical system files (review scripts that perform 'execution' after approval). 5) Limit filesystem permissions for the directory the skill uses so it cannot accidentally overwrite global agent configs. If you cannot audit the setup & execution scripts yourself, avoid enabling cron/automatic runs or limit the skill to manual invocations only.

Self-Improve Framework

A pluggable self-improvement framework for AI agents. Automatically learns from mistakes, corrections, and feedback to continuously improve execution quality.

Description

Self-Improve enables your agent team to evolve over time:

• Scans agent memory logs for learning signals
• Extracts reusable experience rules
• Proposes improvements to system files (with approval workflow)
• Maintains a 3-tier memory system (HOT/WARM/COLD)

When to Use

• Automatic (Cron): Runs every 3 days by default
• Manual trigger: When user asks to "run self-improve" or "learn and improve"
• After significant events: User can request immediate run after major corrections

Installation

clawhub install self-improve

Or with OpenClaw CLI:

openclaw skills install self-improve

Quick Start

1. Configure Paths

Edit user-config.yaml:

storage:
  root: "/path/to/self-improve"
  knowledge_root: "/path/to/learned"
  workspace_root: "/path/to/.openclaw"

owner:
  name: "YourName"
  timezone: "Asia/Shanghai"

2. Run Setup

node scripts/setup.mjs --config user-config.yaml

3. Approve Cron Task

Check proposals/PENDING.md for the suggested Cron task.

How It Works

Scan memory logs -> Extract signals -> Classify by theme
       |
Promote/demote rules between memory tiers
       |
Propose outputs:
  -> System file changes (needs approval)
  -> Knowledge base entries
  -> Blog drafts / methodologies

Memory Tiers

Tier	Location	Purpose
HOT	data/hot.md	Frequently used rules
WARM	data/themes/	Theme-based rules
COLD	data/archive/	Archived rules

Dependencies

• OpenClaw >= 2026.3.0
• Node.js >= 18.0.0

License

MIT License