wema-bmc
Fail
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [CREDENTIALS_UNSAFE]: Hardcoded cleartext credentials for a production service account are embedded directly in the skill file.
- Evidence in
SKILL.md:Username: zedai@wemabank.com,Password: Owzre !Er,s3-37. - The credentials are also hardcoded in the provided Python helper script:
PASS = "Owzre !Er,s3-37". - [COMMAND_EXECUTION]: The skill provides a Python helper script that utilizes
subprocess.runto execute externalcurlcommands. - Evidence in
SKILL.md: Theget_token,query, andadd_worklogfunctions in the Python script all usesubprocess.run(["curl", ...])to interact with the API. - [DATA_EXFILTRATION]: The skill provides ready-to-use commands and script logic designed to query and export sensitive organizational data and Personally Identifiable Information (PII).
- Evidence in
SKILL.md: Detailed examples for querying theCTM:Peopleform which contains names, emails, phone numbers, and department details for bank staff. - Evidence in
SKILL.md: Instructions for queryingHPD:Help Desk,CHG:Infrastructure Change, and other forms containing sensitive internal business processes and incident details.
Recommendations
- AI detected serious security threats
Audit Metadata