T09 · Insecure Skill Coding Practices
- Location
scripts/brand_commercial_os_core.py:685- Finding
Unsupported Marketing Claims and Fabricated Testimonials Bypass Compliance Controls
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is a coherent marketing and negotiation generator, but it can create unsupported commercial claims while presenting its checks as passed and its output as ready to publish.
Treat generated materials as drafts only. Before using this skill for public posts, partner pitches, proposals, or pricing discussions, verify every certification, performance number, testimonial, comparison, and customer claim against real evidence and get human legal/brand review.
scripts/brand_commercial_os_core.py:685Unsupported Marketing Claims and Fabricated Testimonials Bypass Compliance Controls
The skill declares no explicit tool scope or permission boundaries, yet the referenced implementation indicates file-write and network-capable behavior. In an agent setting, undocumented capabilities increase the risk of over-privileged execution, unintended outbound data transfer, or writes based on user-controlled input, especially because this skill is designed as a top-level orchestrator that may handle brand materials and negotiation artifacts.
The trigger phrases are broad enough to match ordinary business-assistance requests, which can cause the skill to activate when the user did not intend to invoke this specific workflow. In a skill with orchestrated generation of pricing, content, and negotiation materials, accidental invocation can expose sensitive business context to unnecessary processing and produce unauthorized outputs or actions.
The FULL_CHAIN flow says N2 runs C → B → A, but the earlier routing rules require B to read A's brand_profile and prefer C's knowledge pack first. This contradiction can cause the content generator to run before the brand fact layer exists, leading to fabricated or stale claims in negotiation and publication outputs despite the skill's own requirement not to invent brand facts.
The guidance says generated cross-platform content can be directly published, but this skill produces brand and marketing material that may contain inaccuracies, compliance issues, unverifiable claims, or platform-policy violations if not reviewed. The surrounding context makes this more dangerous because the outputs are intended for external publication and commercial negotiation, where false or noncompliant statements can create legal, reputational, and contractual harm.
This file forces a specific language/locale in its headings, field labels, and placeholders, and there is no indication that users may opt into another language. Under the policy, a skill artifact that mandates one language without user choice or a stated region-specific rationale is a natural-language policy concern.
This JSON template defines all user-facing question and answer text in Chinese, with no indication that other languages are supported or that the user can choose a locale. Under the policy rule, a skill that implicitly enforces a specific language without opt-in is a natural-language policy violation unless the locale restriction is explicitly justified.
The document is entirely framed as a Chinese "GEO标准答案模板" and all required examples, headings, and output structures are presented only in Chinese. This imposes a specific language/locale on the skill's output without any user opt-in or stated region-specific justification, which matches the language policy violation criteria.
This JSON template contains all user-facing names, descriptions, conditions, risks, and negotiation tips in Chinese, which can impose a specific language on downstream users without opt-in. The policy explicitly flags language or locale constraints when the skill does not offer a choice or clearly justify the locale restriction.
The module description and many user-facing strings throughout the file are written in Chinese, and the skill does not provide any mechanism for the user to select a preferred language or locale. This can violate language/locale policy when a skill implicitly forces one language rather than offering choice or documenting a justified regional constraint.
The Consistency QA step claims that problematic content must be corrected before final output, but the code only logs issues and continues returning the original content bundle unchanged. Because this skill generates external-facing marketing and negotiation materials, prohibited claims or terminology conflicts can propagate into deliverables and be used operationally despite detection.
The one-pager says data feedback commitments will be non-fabricated, yet it includes outcome-oriented assurances such as maximizing investment return without grounding them in supplied evidence or measurable limits. In a commercial negotiation context, this can mislead counterparties and encourage unsupported performance claims that create legal, reputational, or contractual risk.
The Quality Gate is documented as a blocking control that should send failed outputs back for correction, but the implementation only records check results and still returns the final deliverables. In this skill, that means content with completeness or compliance problems can be emitted as if it passed validation, creating a false sense of safety and enabling downstream publication or negotiation use of flawed material.
At L2291-L2293, the code saves execution results to brand_commercial_os_output.json. Although there is a print after saving, there is no prior confirmation prompt or warning in the code comments/docstring that running the script will create or overwrite a local file.