Caveat verdict
browser-harness
browser-harness-skill
Browser automation skill that connects to the user's existing Chrome via CDP — explicitly prohibits storing cookies/tokens/session IDs in files, enforces sensitive-deny mode for banks/email/admin, and operates transparently within the user's own browser session.
⚠ 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 (7)
Recursive delete from root or home — destructive command
setup.md · code · rm -rf ~
Pipe to python — executes piped content as Python code
reference.md · prose · downgraded · | Python
HTTP request to bare IP address — common in malicious payloads
reference.md · code · http://127.0.0.1
Pipe-to-shell pattern (curl | sh) — supply chain attack vector
scripts/run.sh · prose · downgraded · curl -LsSf https://astral.sh/uv/install.sh | sh
Pipe to sh — executes piped content as shell commands
scripts/run.sh · prose · downgraded · | sh
Accesses shell history/config
setup.md · code · ~/.zshrc
Opens WebSocket connection
SKILL.md · prose · downgraded · WebSocket
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class A/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.