Automatically records experience, analyzes skill gaps, updates capability scores, and generates priority tasks to ensure continuous improvement after each task.

This skill is likely doing what it claims, but exercise caution before enabling it automatically. Before running: (1) Inspect evolve.sh yourself — it creates files under /root/.openclaw/workspace (tasks, memory, skills) and will generate P1 tasks; ensure those paths are acceptable. (2) Run the script in a non-production or sandboxed environment first to confirm behavior. (3) Ensure necessary binaries (jq, bc, awk) exist or edit the script to avoid unmet dependencies. (4) If you want least privilege, change TASKS_DIR/MEMORY_DIR to a per-skill sandbox (not /root) so it cannot access or confuse other skills' data. (5) Consider requiring explicit user invocation (not autonomous) until you trust its behavior and avoid letting it auto-enqueue high-priority tasks.

Type: OpenClaw Skill Name: continuous-evolution Version: 1.0.0 The skill bundle implements a 'Continuous Evolution' system that allows the agent to self-evaluate and generate new tasks. A critical shell injection vulnerability exists in `evolve.sh` because it uses unquoted heredocs (`<< EOF`) to write variables like `$task` and `$result` to files, which allows for arbitrary command execution if these variables contain shell substitutions (e.g., `$(...)`). While the 'self-evolution' logic is presented as a feature for autonomous improvement, the combination of insecure script execution and the ability for the agent to programmatically inject high-priority tasks into its own queue poses a significant security risk.