Caveat verdict
soul-memory
Skill is a local long-term memory system for AI agents using vector search; no network-out capabilities are declared and the content describes purely local file-based memory management.
⚠ 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.
Findings (12)
Recursive delete from root or home — destructive command
tests/test_memory_strengthening.py · code · rm -rf /
Pipe to python — executes piped content as Python code
UPGRADE_PLAN_v3.4.md · prose · downgraded · | Python
Accesses OpenClaw config/secrets directly
SKILL.md · prose · downgraded · ~/.openclaw/openclaw.json
References agent memory files
HEARTBEAT.md · code · MEMORY.md
References child_process — can spawn system processes
install.sh · prose · downgraded · child_process
Accesses shell history/config
install.sh · prose · downgraded · ~/.zshrc
References SSH/GPG private keys
keyword_mapping_v3_3.py · prose · downgraded · ssh_key
Opens WebSocket connection
UPGRADE_PLAN_v3.4.md · code · WebSocket
subprocess execution — runs system commands from Python
modules/version_control.py · prose · downgraded · subprocess.run(
Sets world-executable permissions
V3_3_UPGRADE.md · code · chmod 755
Python os.environ.get — reads environment variable
heartbeat-trigger_v3_3.py · prose · downgraded · os.environ.get(
pip3 install — installs Python packages at runtime
install.sh · prose · downgraded · pip3 install
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class D/F). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
No declared permissions — minimal attack surface.
Is this flag fair?
Thanks — recorded.