Agent Self Improve
v0.1.0Analyzes agent performance and optimizes prompts, parameters, workflows, and strategies for continuous self-improvement.
⭐ 0· 159·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description match the implemented modules: a PerformanceAnalyzer and a StrategyOptimizer for prompts, parameters, and workflows. One minor mismatch: package.json mentions "code self-rewriting" but the provided code does not implement any file-modifying or self-rewriting behavior — this appears to be an overstated description rather than hidden functionality.
Instruction Scope
SKILL.md only instructs npm install, shows creating SelfImprovementSystem and calling analyze/improve. The runtime example calls agent.process(input) indirectly (the analyze API expects a function), so the skill itself does not read files, env vars, or send data externally. Users should note that any data passed into analyze/improve comes from the host agent and could include secrets if the agent provides them.
Install Mechanism
No install spec in the registry and SKILL.md suggests running npm install for the package sources provided. package.json has no external dependencies and there are no downloads from untrusted URLs or extraction steps. Installing the package is low-risk.
Credentials
The skill declares no required environment variables, credentials, or config paths. The code only uses standard Node APIs (perf_hooks, process.memory/cpu) and no external tokens or secrets are requested.
Persistence & Privilege
The skill is not forced-always, is user-invocable, and does not modify other skills or system configuration. It does not request elevated or persistent privileges.
Assessment
This skill appears internally consistent and low-risk: review and run its tests locally (npm test) before using it in production, and avoid passing sensitive secrets into the analyze/improve callbacks (e.g., do not pass raw credentials or private data as test inputs). Note the package description mentions "code self-rewriting" but no such behavior exists in the provided files — if you plan to use an auto-updating or self-modifying variant in future, audit any file-write or network code carefully. If you want extra caution, run the skill in an isolated environment or sandbox when first integrating it with a live agent.Like a lobster shell, security has layers — review code before you run it.
latestvk97ae1cr0eqn46km8zbp377hx983ygyj
License
MIT-0
Free to use, modify, and redistribute. No attribution required.