Security audit
Security checks for vulnerabilities and agentic risk
The skill appears to be a documentation-style medical-device compliance helper, but it asks for broader execution, write, and web access than its stated purpose clearly justifies.
Install only if you are comfortable with a regulatory decision-support prompt that currently declares broad Bash, Write, and WebFetch authority. Prefer a revised version that removes or tightly scopes those tools, includes the actual rule library or approved official sources, and clearly states that outputs require qualified human and regulatory review.
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
# non-clinical-checker · 医疗器械非临床研究判定器
[](https://github.com/zhaoxinghua09-cell/lgd-theory)
The manifest presents the skill as a zero-dependency, documentation-style checker with a built-in rule library, yet it grants WebFetch, Bash, and Write capabilities. This mismatch expands the attack surface and can mislead operators into approving a tool with materially more power than its documented purpose requires.
Remote web access is inconsistent with the claim that the system relies on a built-in official rule library and has zero dependencies. WebFetch enables external data exfiltration, untrusted content retrieval, and silent drift from approved rule sources, which is especially risky for compliance-oriented outputs.
Remote web access is inconsistent with the claim that the system relies on a built-in official rule library and has zero dependencies. WebFetch enables external data exfiltration, untrusted content retrieval, and silent drift from approved rule sources, which is especially risky for compliance-oriented outputs.
The documentation states there are no scripts to run and describes the tool as zero-dependency, while the manifest authorizes local execution and file/API-style operations. This inconsistency can cause reviewers and users to underestimate operational risk, making unsafe deployment or over-trust more likely.
The skill produces compliance-style determinations for medical-device non-clinical research but does not clearly warn that outputs are decision support only and should not be treated as authoritative regulatory or product-safety advice. In a medtech context, overreliance on such outputs could lead to flawed study planning, invalid evidence packages, or unsafe regulatory decisions.
SQP-3 applies to natural-language policy violations across all file types, including metadata fields. The file includes a Chinese display name while the skill is otherwise bilingual, but it does not state any user-selectable language preference or locale choice for presentation.