T09 · Insecure Skill Coding Practices
Warning
- Location
- references/03-components-syntax.md:68
- Finding
- Generation of an Unverified Corporate Security Attestation## Vulnerability Details **File Location**: `references/03-components-syntax.md`, lines 68–73 **Vulnerability Type**: Unverified compliance and security claim embedded in a reusable output template **Risk Level**: Medium **Relevant snippet** (English translation of the source): ```markdown ### Type I - Security/Trust Banner - **Format**: `🔒 [Security statement/compliance label/trust mark]` - **Style**: Lock/shield icon, serious but not frightening - **Applicable to**: Security compliance/risk control/privacy statements/closing trust loop - **Example**: `🔒 All the above standards have passed the Ping An Group security review` ``` ### Technical Analysis The reusable Type I presentation component includes a concrete assertion that material has passed a Ping An Group security review. The template does not require the user to supply evidence, identify the scope or date of the review, or confirm authorization to use the corporate endorsement. Because the Skill routes security, compliance, privacy, and trust-oriented pages toward this component, the example may be reproduced or adapted as if it were a verified fact. The surrounding rules also encourage authoritative and trust-building language, increasing the likelihood that generated material will present the statement as an actual attestation rather than illustrative placeholder text. This is a content-integrity weakness rather than a code-execution vulnerability. It does not grant filesystem access, network access, elevated privileges, or control over the host. Its security relevance is the possibility of producing a false or unsupported representation of formal corporate review. ### Attack Path 1. A user requests prompts for a security, privacy, compliance, risk-control, or trust-oriented presentation. 2. The Skill selects or recommends the Type I security/trust banner. 3. The model uses the supplied Ping An Group security-review statement as an output example. 4. Th ...[truncated 863 chars]
- Remediation
- ## Remediation Suggestions 1. Replace the concrete corporate assertion with a neutral placeholder: ```markdown - **Example**: `🔒 [Insert verified compliance status, approving organization, scope, and source]` ``` 2. Add an explicit validation rule stating that approval, audit, certification, compliance, and security-review claims may only be generated when the user provides: - The approving organization - The precise assessment or certification - The reviewed scope - The approval date or validity period - A verifiable source or internal reference - Authorization to display the claim 3. Require uncertain statements to be labeled clearly as proposed, pending, illustrative, or target-state content. 4. Prohibit the model from inferring endorsement from the selected branding track, presentation style, or organization name. 5. Add a quality-control check that rejects unsupported claims containing phrases such as “security approved,” “certified,” “audited,” “compliant,” or “review passed.” 6. Keep trust banners factual and scoped. For example: ```markdown 🔒 Security review status: Pending verification ``` 7. Add provenance metadata or a visible citation to every verified security or compliance assertion included in generated presentation content.