Self Improving Agent V2
PassAudited by VirusTotal on May 11, 2026.
Overview
Type: OpenClaw Skill Name: self-improving-agent-v2 Version: 1.0.0 The skill bundle implements a 'self-improvement' framework designed to help AI agents log errors, user corrections, and new patterns into local markdown files (.learnings/) and workspace memory (SOUL.md, AGENTS.md). The included bash scripts (activator.sh, error-detector.sh, extract-skill.sh) and OpenClaw hooks (handler.js/ts) are standard utility scripts for context injection and file scaffolding; they contain appropriate path validation to prevent traversal and show no signs of data exfiltration, unauthorized network access, or malicious intent.
Findings (0)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Incorrect, sensitive, or overly broad learnings could persist and be reused in later sessions.
The skill explicitly encourages promoting captured learnings into persistent prompt or instruction files that can shape future agent behavior.
Broadly applicable learning | Promote to `CLAUDE.md`, `AGENTS.md`, and/or `.github/copilot-instructions.md`
Review learning entries before promotion, avoid storing secrets, and require user confirmation before changing persistent instruction files.
If enabled, these scripts run automatically in the agent environment and add reminder text to context.
The optional hook setup runs local shell scripts on agent events such as prompt submission or post-tool use. The included scripts appear limited to printing reminders, but enabling hooks creates automatic execution points.
"command": "./skills/self-improvement/scripts/activator.sh"
Enable hooks only if desired, inspect the scripts first, and disable them if the automatic reminders are unnecessary.
Cross-session sharing could expose prior conversation details to another session if used carelessly.
The skill documents OpenClaw tools for reading other sessions and sending learnings across sessions. This is disclosed as part of the workflow, but transcripts may contain sensitive context.
sessions_history — Read another session's transcript
Share only the specific learning needed, avoid forwarding sensitive transcript content, and confirm the target session before sending.
Installing from a moving branch could fetch code different from the reviewed artifacts.
The documentation includes a manual install from an unpinned GitHub repository, while the evaluated registry entry has unknown source/homepage metadata. This is a provenance note, not evidence of malicious code.
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
Prefer the reviewed registry package, or pin and inspect any GitHub revision before installing manually.