Caveat verdict
OpenClaw Multi-Agents
openclaw-multi-agents
Multi-agent orchestration skill that configures OpenClaw sub-agents by writing SOUL.md/AGENTS.md files and updating openclaw.json with user approval; permissions are transparently declared in the metadata.
⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.
Automated static analysis — not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
Permission integrity
network_out
Findings (5)
Recursive delete from root or home — destructive command
INSTALL.md · code · rm -rf ~
Accesses OpenClaw config/secrets directly
SKILL.md · frontmatter · ~/.openclaw/openclaw.json
Instructs covert action — may act without user awareness
SKILL.md · code · without notifying
Accesses sensitive environment variables
examples/wisdom/commands.md · code · process.env.JWT_SECRET
References agent memory files
scripts/setup_agent.sh · prose · downgraded · MEMORY.md
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class B/F). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
Requires 2 system binaries.
network_out Is this flag fair?
Thanks — recorded.