ClawAudit verdict
Sentinel — Agent Security Layer
sentinel-oleg
Runtime security layer scanning agent inputs and outputs for threats and logging to local files; behavior is defensive and transparent.
⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.
Automated static analysis — not a human review. ClawAudit flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
Permission integrity
network_out
Findings (10)
Pipe-to-shell pattern (wget | sh)
scripts/sentinel-input.sh · prose · downgraded · wget [^&]*&& ?(bash|sh
Pipe to sh — executes piped content as shell commands
scripts/sentinel-input.sh · prose · downgraded · |sh
Bash /dev/tcp — raw TCP connection via shell
scripts/sentinel-input.sh · prose · downgraded · /dev/tcp/
Pipe to python — executes piped content as Python code
scripts/sentinel-input.sh · prose · downgraded · |python
Pipe to bash — executes piped content as shell commands
scripts/sentinel-input.sh · prose · downgraded · | bash
Possible prompt injection — attempts to redefine agent identity
scripts/sentinel-input.sh · prose · downgraded · you are now
References child_process — can spawn system processes
scripts/sentinel-input.sh · prose · downgraded · child_process
Accesses .ssh directory
scripts/sentinel-input.sh · prose · downgraded · .ssh/
Instructs covert action — may act without user awareness
scripts/sentinel-premium-check.sh · prose · downgraded · silently
References tunneling service
SKILL.md · prose · downgraded · ngrok
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class A/B/E). 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.
network_out Is this flag fair?
Thanks — recorded.