ClawAudit verdict
agentshield
agentshield-audit
Security audit tool that runs 77 tests locally and generates Ed25519 identity keys that stay on device; only pass/fail scores (no prompts or keys) are sent to the API, consistent with its stated privacy-first trust-infrastructure purpose.
⚠ 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
package_install
Findings (14)
Unicode homoglyph detected — uses lookalike characters to evade pattern matching
agentshield_tester.py · prose
References sudo — requests elevated privileges
INSTALLATION.md · code · sudo
Recursive delete from root or home — destructive command
INSTALLATION.md · prose · downgraded · rm -rf ~
Prompt injection — tries to override agent instructions
echoleak_test.py · prose · downgraded · ignore previous instructions
subprocess execution — runs system commands from Python
PLATFORMS.md · code · subprocess.run(
Downloads executables from external URLs
INSTALLATION.md · code · Download from https://
apt-get install — installs system packages
INSTALLATION.md · code · apt-get install
pip3 install — installs Python packages at runtime
SKILL.md · code · pip3 install
Base64 encoding/decoding
SKILL.md · code · base64_encode
References agent configuration files
QUICKSTART.md · code · agent_config
Accesses .ssh directory
agentshield_tester.py · prose · downgraded · .ssh/
Python os.environ.get — reads environment variable
audit_client.py · prose · downgraded · os.environ.get(
Python os.getenv — reads environment variable
complete_handshake.py · prose · downgraded · os.getenv(
Popular HTTP library — network access
supply_chain_scanner.py · prose · downgraded · Got
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class A/B/D). 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.
package_installnetwork_indata_encoding Is this flag fair?
Thanks — recorded.