cspm-casb-attack
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill documentation provides instructions for downloading and executing scripts directly from external repositories without prior verification steps. Evidence includes the use of piped shell execution:
curl -s https://raw.githubusercontent.com/kubescape/kubescape/master/install.sh | /bin/bashinguides/cspm-casb-attack-playbook.md.\n- [DATA_EXFILTRATION]: The skill contains specific procedures for harvesting hardcoded credentials and sensitive metadata from Terraform state files. Evidence includes instructions to use secret-scanning tools liketrufflehogandjqfilters on local and remote state files inpayloads.mdandguides/cspm-casb-attack-playbook.md.\n- [PERSISTENCE]: The skill provides a bash script intended to automate the suppression of new security findings in a continuous loop, facilitating long-term stealth in a compromised environment. Evidence includes thewhile trueloop logic inguides/cspm-casb-attack-playbook.mdthat monitors forWorkflowStatus=NEWand updates it toSUPPRESSED.\n- [COMMAND_EXECUTION]: The skill facilitates the modification of cloud security postures through extensive use of administrative CLI tools. Evidence includes commands to update finding statuses viaaws securityhub batch-update-findingsand dismissing assessments usingaz security assessment create.\n- [DYNAMIC_EXECUTION]: The skill utilizes dynamic policy evaluation engines to process and apply security configurations at runtime. Evidence includes the use ofopa evalandkyverno applyfor policy-as-code bypass demonstrations inpayloads.md.\n- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by ingesting and acting upon untrusted data from cloud environments and state files.\n - Ingestion points: The skill reads external data from files like
terraform.tfstate(payloads.md) and tool outputs such asprowler-checks.txt(TC-CP-003).\n - Boundary markers: The instructions lack delimiters or warnings to ignore potentially malicious content within these data sources.\n
- Capability inventory: The skill utilizes powerful command-line utilities including
aws,az,gcloud,kubectl, andcurlacross multiple scripts and guides.\n - Sanitization: No evidence of data sanitization or validation was found before processing these external inputs.
Recommendations
- AI detected serious security threats
Audit Metadata