Caveat verdict
benchclaw - openclaw-benchmark
benchclaw
The skill explicitly discloses it POSTs encrypted per-question aggregates and truncated stdout/stderr to the BenchClaw server; while sanitization is attempted and upload can be disabled, sending agent execution output to a third-party benchmarking API is a genuine concern beyond the stated benchmarking purpose.
⚠ 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
file_read+write
package_install
Findings (8)
References sudo — requests elevated privileges
SKILL.md · code · sudo
Dynamic __import__('os') — Python OS command execution
scripts/utils.py · prose · downgraded · __import__("os")
subprocess execution — runs system commands from Python
scripts/agent_cli.py · prose · downgraded · subprocess.run(
Uses exec() — may execute shell commands
scripts/session.py · prose · downgraded · exec(
Python os.environ.get — reads environment variable
scripts/crypto.py · prose · downgraded · os.environ.get(
Opens WebSocket connection
scripts/main.py · prose · downgraded · WebSocket
Python urllib.request — network access
scripts/main.py · prose · downgraded · urllib.request
Python os.getenv — reads environment variable
scripts/session.py · prose · downgraded · os.getenv(
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class B/D). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
Requires 3 system binaries. (1 elevated: pip).
network_infile_writepackage_install Is this flag fair?
Thanks — recorded.