Strict Self-Improving Agent (Rule of 3)

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...

This skill appears to do what it says — it only injects reminders, detects command errors locally, scaffolds new skills from learnings, and aggregates promotion candidates. Before enabling it: 1) review the scripts (activator.sh, error-detector.sh, extract-skill.sh, promote-review.sh) so you understand what files they write and where; 2) be aware hooks run with the agent's permissions and will create/modify files under ~/.openclaw/workspace and memory/core; 3) logs and learning entries may contain command output or context that could include secrets — decide whether to redact or avoid logging sensitive output; 4) if activations are too frequent, use matcher filters (as documented) to limit when hooks run; and 5) only enable the OpenClaw hook(s) if you trust the skill and want persistent reminders and local file writes. If you want additional assurance, run the scripts in a dry-run or test workspace first.

Type: OpenClaw Skill Name: strict-self-improvement Version: 1.0.2 The skill bundle is classified as suspicious due to a critical shell injection vulnerability found in `scripts/extract-skill.sh`. The `SKILL_NAME` argument is directly interpolated into a command executed by `sed` and `awk` without proper sanitization or quoting, specifically in the line generating the H1 title. This allows an attacker to execute arbitrary shell commands by crafting a malicious `skill-name` (e.g., `my-skill; rm -rf /`). While the script's intended purpose is benign (creating a skill scaffold), this vulnerability allows for unauthorized command execution, making it a significant security risk. Other components, including the agent's self-improvement instructions and hook scripts, appear benign and even include security-positive restrictions like human-in-the-loop approval for promoting knowledge to core agent instructions.