T09 · Insecure Skill Coding Practices
- Location
SKILL.md:35- Finding
Meeting transcripts may be disclosed to a third-party API without explicit consent or data-protection controls
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md, line 35
Vulnerability Type: Unprotected transmission of potentially sensitive meeting data to a third-party service
Risk Level: MediumVulnerable Code Snippet
markdown Use the DeepSeek API for a strict four-round process to ensure information completeness and compliance with the length requirement.The snippet above is an English translation of the instruction at line 35.
Technical Analysis
The skill directs the agent to use the external DeepSeek API when processing meeting transcripts. These transcripts can contain personal information, confidential business discussions, financial figures, technical details, credentials disclosed during conversation, internal decisions, and other regulated or proprietary data.
The documented workflow emphasizes extracting and validating all information, which implies that substantially complete transcript content may be submitted to the external provider. However, the skill does not require:
- Explicit and informed user consent before external transmission.
- Disclosure of the third-party processor.
- Redaction of personal, confidential, or credential-related information.
- Data minimization before submission.
- Validation of the API endpoint and transport security.
- A defined retention, deletion, or training-data policy.
- A local-processing option for sensitive documents.
- Secure credential handling for the API integration.
This is an insecure skill-design practice because data may cross a trust boundary without appropriate safeguards. The repository contains only
SKILL.md; therefore, no implementation was available to verify whether protections exist outside the documented workflow.Attack Path
- A user uploads a PDF or Word document containing a meeting transcript.
- The skill extracts the document's complete textual content.
- Following the instruction at line 35, the agent sends meeting content to the DeepSeek API for mult ...[truncated 1371 chars]
- Remediation
View remediation
Remediation Suggestions
- Require explicit, informed user consent before sending any document content to DeepSeek or another external provider.
- Clearly disclose the provider, categories of transmitted data, processing purpose, retention period, deletion policy, and whether submitted data may be used for model training.
- Add a local-only processing mode and make it the default for documents classified as confidential or regulated.
- Perform data minimization so that only content necessary for each processing stage is transmitted.
- Detect and redact credentials, personal identifiers, financial account details, health information, and other sensitive fields before transmission.
- Allow users to review and approve the redacted payload before it leaves the local environment.
- Require HTTPS with certificate validation and restrict requests to an explicitly allowlisted API endpoint.
- Store API credentials in an approved secret manager or protected environment variable; never place them in skill text, source files, generated documents, or logs.
- Disable sensitive request and response logging, or apply comprehensive redaction before logging.
- Establish contractual and technical retention controls, including deletion guarantees and restrictions on secondary use.
- Document an incident-response process for unintended disclosure.
- Update the workflow to stop processing when consent is absent or when the document's sensitivity exceeds the configured external-processing policy.