T09 · Insecure Skill Coding Practices
Warning
- Location
- scripts/screen_aml.py:34
- Finding
- Sensitive AML identity data may be retained by the remote provider without explicit opt-in<![CDATA[ ## Vulnerability Details **File Location**: `scripts/screen_aml.py:34-47` **Related Documentation**: `SKILL.md:86-87` **Vulnerability Type**: Privacy-unsafe remote data retention default **Risk Level**: Medium ### Complete Code Snippet ```python def screen_aml(full_name: str, date_of_birth: str = None, nationality: str = None, document_number: str = None, entity_type: str = "person", threshold: int = None, vendor_data: str = None) -> dict: api_key = get_api_key() payload = {"full_name": full_name, "entity_type": entity_type} if date_of_birth: payload["date_of_birth"] = date_of_birth if nationality: payload["nationality"] = nationality if document_number: payload["document_number"] = document_number if threshold is not None: payload["aml_match_score_threshold"] = threshold if vendor_data: payload["vendor_data"] = vendor_data r = requests.post(ENDPOINT, headers={"x-api-key": api_key, "Content-Type": "application/json"}, json=payload, timeout=60) ``` The documented API behavior is: ```text save_api_request | boolean | No | true | Save in Business Console ``` ### Technical Analysis The script legitimately needs to send the screening subject's identity data to the declared Didit AML endpoint. However, it does not set the API's `save_api_request` field and provides no command-line control for it. According to `SKILL.md`, the service defaults this option to `true`. Consequently, names, dates of birth, nationalities, document numbers, and tracking identifiers may be stored in the provider's Business Console even when the user only intends to conduct a one-time screening. Date-of-birth and government-document information are particularly sensitive personal data. The transmission itself is not covert exfiltration: it uses HTTPS, targets the documented service, and is central to the Skill's declared purpose. ...[truncated 1528 chars]
- Remediation
- <![CDATA[ ## Remediation Suggestions 1. Set `save_api_request` to `false` in the request payload by default: ```python payload = { "full_name": full_name, "entity_type": entity_type, "save_api_request": False, } ``` 2. Add an explicit opt-in command-line option such as `--save-api-request`. 3. Only change the payload value to `true` when the user supplies that option. 4. Clearly disclose what data is transmitted, whether it will be retained, and how users can request deletion. 5. Avoid collecting optional identity fields unless they are necessary for the required match accuracy. 6. Document the provider's retention period, access controls, deletion procedure, data-processing role, and applicable regional transfer requirements. 7. Consider requiring confirmation before transmitting a document number or other high-impact identifier. ]]>