← Back to Skills Marketplace
Self Improving Agent
by
lovensky1992-wk
· GitHub ↗
· v1.0.1
· MIT-0
418
Downloads
0
Stars
1
Active Installs
8
Versions
Install in OpenClaw
/install self-improving-learner
Description
错误学习闭环：记录失败和纠正 → Pattern-Key 分类 → 定期复盘 → 整合长期记忆 → 防止再犯。 Use when: (1) 命令/操作意外失败且原因值得记录, (2) 用户纠正了错误（"不对"/"Actually..."/"你搞错了"）, (3) 用户要求的能力不存在（能力缺口信号）, (4) 外...
Usage Guidance
This skill appears to do what it says: remind the agent to log errors and convert repeated issues into persistent learnings. Before installing or enabling hooks globally, note: (1) the activator and error-detector are local scripts that will run with the same permissions as your agent and can write to your workspace (.learnings/, AGENTS.md, etc.); (2) enable the hooks only in the contexts you trust (project-level vs user-level); (3) review scripts (activator.sh, error-detector.sh, extract-skill.sh) and run extract-skill.sh with --dry-run to confirm behavior; (4) if you expect any cross-session operations (sessions_history/sessions_send), explicitly audit any runbooks that instruct the agent to use those APIs because they can access other sessions' data. Finally, be aware of minor metadata/version mismatches in the package (cosmetic) and verify origin if provenance is important.
Capability Analysis
Type: OpenClaw Skill
Name: self-improving-learner
Version: 1.0.1
The skill bundle implements a structured 'self-improvement' loop for AI agents, allowing them to log errors, user corrections, and new insights into local markdown files. It includes shell scripts (activator.sh, error-detector.sh) and OpenClaw hooks (handler.js/ts) that provide passive reminders to the agent without executing harmful commands or exfiltrating data. The skill-extraction script (extract-skill.sh) demonstrates security awareness by explicitly validating paths to prevent directory traversal.
Capability Tags
Capability Assessment
Purpose & Capability
Name/description (self-improvement / learning loop) match the included artifacts: activator + error detector hooks, a hook handler for bootstrap injection, templates, and a helper to extract a learning into a new skill. The files and scripts are proportionate to the stated purpose. Minor metadata/version inconsistencies exist between registry metadata, _meta.json, and README/SKILL.md versions (cosmetic, not functional).
Instruction Scope
SKILL.md instructs the agent to read/write project workspace files (.learnings/, AGENTS.md, SOUL.md, MEMORY.md) and to run periodic reviews — this is expected for a learning-capture skill. The documentation also references OpenClaw session APIs (sessions_history, sessions_send, sessions_spawn) as potential capabilities; those are only documented examples and are not automatically invoked by the included code. If an agent were later instructed to call cross-session APIs, that would expand scope and privacy impact — review any runbooks that authorize cross-session reads/writes before enabling globally.
Install Mechanism
There is no remote install/download step in the skill bundle provided. All included scripts and hook handlers are local files (no network fetches or extracts). This is a low-risk install profile. The extract-skill.sh helper writes files under the current working directory when run (it validates and disallows absolute paths and '..' traversal).
Credentials
The skill declares no required env vars or credentials — appropriate. One included helper (scripts/error-detector.sh) reads CLAUDE_TOOL_OUTPUT to detect errors; this is a platform-provided environment variable and aligns with the purpose, but it is not listed in requires.env (document-only). There are no requests for secrets or unrelated cloud credentials.
Persistence & Privilege
always:false and hooks are opt-in. The hook handler injects a virtual bootstrap file (SELF_IMPROVEMENT_REMINDER.md) on agent:bootstrap; other scripts are run only if the user configures hooks (references/hooks-setup.md). The skill does not modify other skills' configuration or request permanent system-level privileges.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install self-improving-learner - After installation, invoke the skill by name or use
/self-improving-learner - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
Daily sync
v1.0.14
Daily sync
v1.0.13
Daily sync
v1.0.12
Daily sync
v3.0.7
Daily sync
v3.0.6
Daily sync
v3.0.5
Daily sync - fix slug conflict
v1.0.0
Initial release: Fork of pskoett/self-improving-agent with Chinese localization and operational improvements
Metadata
Frequently Asked Questions
What is Self Improving Agent?
错误学习闭环：记录失败和纠正 → Pattern-Key 分类 → 定期复盘 → 整合长期记忆 → 防止再犯。 Use when: (1) 命令/操作意外失败且原因值得记录, (2) 用户纠正了错误（"不对"/"Actually..."/"你搞错了"）, (3) 用户要求的能力不存在（能力缺口信号）, (4) 外... It is an AI Agent Skill for Claude Code / OpenClaw, with 418 downloads so far.
How do I install Self Improving Agent?
Run "/install self-improving-learner" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Self Improving Agent free?
Yes, Self Improving Agent is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Self Improving Agent support?
Self Improving Agent is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Self Improving Agent?
It is built and maintained by lovensky1992-wk (@lovensky1992-wk); the current version is v1.0.1.
More Skills