Security audit
Security checks for vulnerabilities and agentic risk
This sales-coaching skill is not malicious, but it asks the agent to read and persist sensitive customer and meeting material with broad triggers and limited user-control boundaries.
Install only if you are comfortable with the agent searching the specified local knowledge-base folders, reading full customer or meeting transcripts, inferring stakeholder motives and relationships, and saving detailed reports locally. Prefer invoking it only for explicit customer-analysis tasks and review or disable the save-report step when handling confidential, investor, or internal operations material.
The skill also claims investment-analysis capability, which is materially outside the declared sales-diagnosis purpose. This unjustified expansion increases risk because the agent may process highly sensitive financing, valuation, or investor information without users understanding they invoked a broader analysis function.
文件整体以中文固定描述技能行为与输出，面向“全球Top 10的销售专家”的通用分析场景，但未说明是否可根据用户偏好切换语言。按规则，若技能隐含强制特定语言且无用户选择或正当的地区限定说明，可构成语言/locale 策略问题。
The trigger list contains broad everyday phrases such as meeting analysis and sales capability terms, making accidental invocation likely. Overbroad activation is dangerous here because the skill is authorized to read local files, inspect transcripts, infer sensitive attributes, and save reports, so an unintended trigger can lead to unnecessary access and data processing.
The instructions explicitly direct the agent to read local customer profiles and full transcript contents, then save an analysis report, but the skill does not transparently disclose these read/write behaviors to the user. This is dangerous because it enables access to and propagation of potentially confidential personal, commercial, and strategic data without informed consent or least-privilege safeguards.
The skill instructs the agent to read customer archives and full meeting/transcript content, infer role relationships, attitudes, personal incentives, and concerns, and then save a detailed report. This creates a concentrated privacy and confidentiality risk because the agent is directed to process and persist sensitive personal and business intelligence far beyond minimal analysis, with no consent flow, minimization, or retention controls.
The skill is presented as an analysis/coaching tool, but it also instructs the agent to persist generated reports into a local filesystem path. This creates an undeclared side effect: the skill can write sensitive business analysis derived from customer files and transcripts into a knowledge base without an explicit user confirmation step.
The example invocations in the trigger section are ambiguous and lack boundaries, which increases the probability the skill activates during ordinary discussion. In this skill context, accidental activation is more serious because it can initiate analysis of customer records and meeting content without a clearly informed request.
The skill expands from sales diagnosis into internal meeting task tracking and decision recording, which broadens behavior beyond the declared purpose. While not inherently dangerous on its own, this kind of scope creep increases the chance of collecting, restructuring, and persisting unrelated internal operational data under an ambiguously triggered skill.
No suspicious patterns detected.
