Security audit
Security checks for vulnerabilities and agentic risk
The skill is a routing optimization guide with no executable installer, hidden network behavior, persistence, or automatic authority, though users should be careful about the routing logs they provide.
Before installing, confirm where dispatch logs come from and keep them minimized or redacted, especially if they include prompts, user identifiers, business data, or secrets. Review any generated routing-rule changes manually.
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
stats = rule_stats[rule_name] stats['success_rate'] = stats['success'] / stats['hits'] if stats['hits'] > 0 else 0
return rule_stats
**输出示例**:
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
stats = rule_stats[rule_name] stats['success_rate'] = stats['success'] / stats['hits'] if stats['hits'] > 0 else 0
return rule_stats
**输出示例**:
The skill explicitly discusses collecting routing logs, success data, and user satisfaction signals, but provides no privacy, minimization, retention, consent, or sanitization guidance. In practice, dispatch logs and satisfaction metadata can contain sensitive user prompts, task content, identifiers, or behavioral data, creating a meaningful privacy and data-governance risk.
The inline comments say that high-success rules should have their priority lowered for faster matching and low-success rules should have their priority raised to give them more chances. But the code decreases the numeric priority for high-success rules and increases it for low-success rules, then sorts ascending by adjusted_priority, which makes low numeric values match earlier; this directly contradicts the stated rationale for the low-success branch.
The roadmap explicitly says '添加更多中文关键词' and the examples/config emphasize Chinese keywords, but the document does not state that the skill is China-specific or that users can choose another language/locale. This can be a natural-language policy issue because it steers behavior toward a specific language without opt-in or documented justification.
The configuration uses simple keyword matching with broad, high-frequency terms such as 'code', 'write', 'design', 'learn', and 'task' across multiple roles. This can cause unintended skill routing, over-collection of user intent signals, and misclassification of normal conversation, especially because the trigger set spans both English and Chinese without contextual disambiguation.
The config hard-codes multilingual keyword policy that influences behavior based on language-specific terms, but there is no indication of user consent, disclosure, or a way to opt out. This can create quiet profiling or routing behavior that users do not expect, particularly for bilingual users whose normal wording may trigger specialized handling.
No suspicious patterns detected.