fintech-compliance
Warn
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/audit-evidence-collector.shperforms extensive automated execution of dozens of cloud and system CLI commands, includingaws,az,gcloud,psql,trivy, andnpmto extract environment details. - [DATA_EXFILTRATION]: The automated evidence collector in
scripts/audit-evidence-collector.shextracts and bundles highly sensitive security metadata into a single compressed archive (.tar.gz). Collected data includes IAM credential reports (user security postures), firewall rules (NSGs/ACLs), VPC flow logs, and local authentication logs from/var/log/app/auth.log. While not transmitted externally by the script, this behavior performs the 'collection' and 'staging' phases of a data exfiltration attack. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in its evidence collection routine.
- Ingestion points: Processes untrusted external data from system authentication logs, git commit history, and container vulnerability reports.
- Boundary markers: None implemented to distinguish between system data and potentially injected instructions within logs or commit messages.
- Capability inventory: Extensive subprocess calls for system configuration and file writing across all cloud providers.
- Sanitization: No sanitization or filtering of ingested log content is performed prior to archiving and potential future analysis by an agent.
Audit Metadata