continuous-monitoring

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). Yes—runtime monitoring scripts fetch outsider-authored free text from public web sources (e.g., curl to https://crt.sh/?... and curl -s "$JS_URL" for arbitrary JS URLs, plus GitHub org events via gh api), then embed the resulting extracted content (subdomain lists/endpoints/findings) into notification messages that are sent to the LLM agent context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The GitHub module URLs (e.g., go install github.com/projectdiscovery/subfinder/v2/cmd/subfinder@latest and similar lines) are invoked in the GitHub Actions runtime to fetch and install remote binaries (executing remote code), which the workflow depends on.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs creating a systemd service file under /etc/systemd/system and running "sudo systemctl enable --now", which requires elevated privileges and modifies system state (and also includes cron entries that write to /var/log), so it pushes changes that can compromise the host.