OpenClaw Agent Orchestrator

Create and repair durable OpenClaw agents, bindings, and runtime state. Use when the user wants real subagents, durable agent entries, or proof that orchestr...

This skill appears to do what it claims, but it will modify your local OpenClaw workspaces and call the openclaw CLI. Before using it: (1) ensure the openclaw CLI is installed and you understand its permissions; (2) back up any existing workspace you care about because create-agent.sh will remove an existing target workspace for the agent-id; (3) confirm OPENCLAW_HOME if you don't want ~/.openclaw used; (4) run scripts/verify-runtime.sh first to inspect current state; and (5) review the scripts yourself (they are short and included) before granting the agent permission to run them.

Type: OpenClaw Skill Name: openclaw-agent-orchestrator Version: 0.6.0 The skill bundle contains a path traversal vulnerability in `scripts/create-agent.sh`. The script uses `shutil.rmtree()` on a target directory path constructed directly from the `AGENT_ID` argument without sanitization, which could allow for the deletion of arbitrary directories if the agent is manipulated into using a malicious ID (e.g., `../../`). While the instructions in `SKILL.md` and the logic in `scripts/verify-runtime.sh` appear aligned with the stated purpose of agent orchestration, the lack of input validation on destructive file operations is a significant security risk.