Security audit
Security checks for vulnerabilities and agentic risk
This skill searches public company information, but it asks the agent to omit sources in reports and uses a local search helper, so it should be reviewed before use.
Review before installing. The skill is aimed at public company research, but it may send searched company names and business-intelligence intent to a Baidu search helper, and its main instructions tell the agent not to show sources. Ask the agent to include source links and collection dates, verify important claims against official registries, and do not rely on it alone for legal, financial, or reputational decisions.
The skill invokes a shell command to run another skill's Python script, but the manifest does not declare any explicit tool scope or allowed-tools boundary. This weakens containment and reviewability, making it easier for the skill to gain broader execution capability than users or the platform may expect.
The skill explicitly instructs the agent to present collected business information without citing sources, even though the workflow depends on aggregating public data and recommends verification. Removing provenance prevents users from assessing reliability, detecting errors, or distinguishing official records from third-party summaries, which can mislead due-diligence or compliance decisions.
Presenting collected business intelligence without source citations reduces transparency into where the information came from and whether it is trustworthy. In a due-diligence context this can enable silent propagation of stale, incorrect, or selectively framed data while appearing authoritative.
An instruction to omit sources from the final report facilitates unattributed disclosure of aggregated information and obscures whether data was drawn from authoritative public registries, commercial databases, or news reports. This increases the chance of unsafe reliance on unverified or context-stripped information, especially when reporting legal or reputational risk data about companies or individuals.
This script sends company names and derived business-intelligence queries to an external search component without explicit notice or consent at runtime. In due-diligence or competitive-analysis contexts, the queried target itself can reveal sensitive business intent, so silent transmission creates a real privacy and operational-security exposure.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
]
try:
result = subprocess.run(
cmd,
capture_output=True,
text=True,
Natural-language policy violations include forcing a specific language without user opt-in. This skill's description and operating instructions are entirely in Chinese, and there is no indication that users may interact in another language or that the locale limitation is intentional and documented.
The entire skill content is presented only in Chinese, beginning with the Chinese title and continuing throughout the instructions and examples. Under the policy rule, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is explicitly documented and justified, which is not present here.
User-facing documentation, query templates, status messages, and report output are primarily hardcoded in Chinese. There is no indication that users can opt into another language or that the Chinese-only behavior is a justified regional constraint.
No suspicious patterns detected.