Security audit
Security checks for vulnerabilities and agentic risk
This is a coherent obstetrics and gynecology reference package, but it includes high-stakes dosing, surgery, emergency, and pregnancy-related instructions without consistent clinician-only safeguards.
Review before installing in any general-purpose agent. This package should be limited to clinician-supervised education or professional decision support, with explicit warnings that it is not for self-diagnosis, self-medication, emergency self-management, or standalone treatment decisions. Add global and per-skill guardrails requiring qualified clinician review, current local guideline verification, contraindication screening, and urgent in-person care escalation for emergencies.
声明描述指向的是一个妇产科临床技能注册表的数据/内容资源，而实际代码实现的是围绕该清单的辅助工具：统计分类、生成日报告文件、并打印报告。也就是说，代码的主要功能不是提供“187项妇产科临床技能注册表”内容本身，而是对一个现有的 SKILL.md 进行分析汇总。虽然主题相关，但用途发生了实质偏移：从临床技能清单变成了清单报告生成器。未见危险或越权行为，但其实际能力（文件读取、统计、生成报告文件）并未在声明中体现，因此应判定为描述与行为不一致。
声明描述的是内容/数据集性质：一个妇产科临床技能注册表；但提供的代码并不体现医学技能注册表内容本身，而是一个配套的命令行检索工具。其主要行为是接收关键词、调用 grep 搜索 SKILL.md 中的 name/display/trigger 字段并格式化输出结果。这与声明的主要目的存在实质差异：描述强调医学技能注册表，代码体现的是搜索实现。虽然搜索可能是支持性功能，但当前代码块本身未表现出注册表内容管理或医学知识功能，因此应判定为描述与行为不一致。
The skill provides concrete hormone regimens, routes, doses, and interpretation steps for a diagnostic medication challenge without clearly stating that this is clinician-only guidance and must not be self-administered. In a medical skill about amenorrhea, omission of that warning materially increases the risk that a user could self-medicate with estrogen/progestogen despite contraindications such as pregnancy, thrombotic risk, liver disease, or structural pathology, leading to misdiagnosis or harm.
The skill gives concrete obstetric surgical timing recommendations and even suggests output phrases such as when to operate, but it does not include any warning that this is medical decision support requiring licensed clinician oversight and patient-specific evaluation. In a medical context, users may over-trust the guidance as prescriptive, creating risk of delayed or inappropriate surgery during pregnancy, with potential harm to both pregnant patient and fetus.
The skill provides direct high-risk medical management instructions for a newborn or fetus, including immediate glucocorticoid treatment and possible surgical referral, without requiring urgent specialist confirmation, emergency evaluation, dosing safeguards, or explicit warnings against autonomous use. In this context, incorrect application could delay differential diagnosis, miss salt-wasting crisis or other disorders of sex development, and contribute to irreversible or controversial interventions such as genital surgery.
This skill gives stage-specific, personalized cervical cancer treatment recommendations, including surgery, radiotherapy, chemotherapy, fertility-preserving options, and postoperative adjuvant therapy, but provides no warning that the output is informational only and cannot replace oncology/gynecology specialist judgment. In a high-stakes cancer context, omission of safety disclaimers and escalation guidance can lead users to rely on the model for definitive treatment decisions, causing delayed care, inappropriate therapy selection, or serious harm.
This skill gives concrete drug selection, dosing, frequency, and duration instructions for pregnant patients and newborns, which is high-risk medical guidance, but it does not include a prominent requirement for clinician verification, contraindication review, diagnostic confirmation, or emergency escalation. In this context, omissions or outdated recommendations could directly cause patient harm because the target populations are especially vulnerable and the output is framed as actionable prescribing guidance.
The skill gives concrete pregnancy-related dosing instructions for vitamin B6, iron, and calcium without requiring clinician review, screening for contraindications, or user-facing warnings about when self-treatment is unsafe. In a pregnancy context, seemingly routine supplements and symptom treatment can mask serious disease, be inappropriate for the patient's condition, or lead to delayed care.
The skill recommends symptom-directed treatment in pregnancy, including laxatives and other interventions, without explaining risks, limits, or urgent-care triggers. This is dangerous because constipation, edema, pain, or hypotension in pregnancy may represent complications that need prompt assessment, and unsupervised treatment can delay diagnosis or expose the patient to inappropriate therapies.
The skill gives direct pregnancy-termination guidance for high-risk congenital heart disease cases without an explicit warning that it is for clinician use only and not a substitute for urgent specialist evaluation. Because it presents threshold-based recommendations in imperative language, a user could act on high-stakes reproductive and cardiac decisions without adequate confirmation, creating serious risk of harm to both pregnant patient and fetus.
The output section operationalizes the triage into fixed conclusions including '不宜妊娠需终止' and adds perioperative advice, but provides no explicit limitation notice, informed-consent context, or requirement for immediate in-person medical supervision. This makes the skill more dangerous because it converts nuanced specialist decision-making into a simple output that could be used by non-specialists or patients as actionable medical direction.
This skill provides concrete ovarian stimulation, monitoring, and ovulation-trigger instructions that can directly affect dosing, timing, and management of OHSS risk, but it does not clearly state that it is not a substitute for licensed reproductive endocrinology judgment. In a high-risk medical context like IVF/ICSI, omission of that warning can encourage over-reliance on the document as an operational protocol, increasing the chance of patient harm from misapplication, incomplete assessment, or delayed escalation.
The skill gives urgent intrapartum intervention steps and escalation to emergency delivery, but it does not explicitly state that the content is decision-support only, must be used by qualified clinicians, and cannot replace institutional emergency protocols or bedside judgment. In a high-acuity obstetric setting, omission of these guardrails can encourage overreliance on simplified instructions, increasing the risk of delayed, inappropriate, or unsafe maternal-fetal management.
The skill gives concrete postoperative cancer risk stratification and adjuvant treatment recommendations, including when to observe, irradiate, or administer chemotherapy, without an explicit warning that the output is only decision support and must not replace oncology/gynecologic specialist judgment. In a high-stakes medical context, omissions, ambiguous staging inputs, or overconfident use by a non-expert could lead to undertreatment or overtreatment with serious patient harm.
The skill provides decision support for invasive fetal interventions and explicitly outputs whether to perform in-utero treatment and which modality to use, but it does not present a user-facing safety warning, emergency boundaries, or a clear requirement that decisions be made only by qualified fetal medicine specialists in an appropriate center. In a high-risk obstetric context, omission of these safeguards can enable unsafe reliance on the skill for medical decision-making, with potential for severe maternal and fetal harm.
This skill gives concrete high-risk obstetric management recommendations about fetal surveillance and delivery timing for FGR, including thresholds tied to AEDV/REDV, ductus venosus abnormalities, and intervention before 34 weeks, but it does not warn that these recommendations require specialist clinical judgment, patient-specific assessment, and adherence to local protocols/NICU capability. In a safety-critical medical context, omission of such guardrails can lead users to over-trust the skill as prescriptive guidance, creating risk of premature delivery, delayed delivery, or inappropriate treatment decisions.
This file provides concrete chemotherapy agents, schedules, and partial dosing information for high-risk oncologic treatment without a prominent warning that it is not for unsupervised use and requires qualified specialist oversight. In the context of an agent skill, users or downstream systems could mistake it for directly actionable treatment instructions, creating risk of severe toxicity, dosing error, organ injury, marrow suppression, or death.
This skill gives concrete diagnostic and treatment guidance for a potentially serious pregnancy complication, including IV support and antiemetic selection, but does not clearly warn that it is not a substitute for clinician judgment or urgent in-person evaluation. In this context, omission of escalation guidance can delay emergency care for dehydration, electrolyte abnormalities, altered mental status, or Wernicke encephalopathy, increasing maternal-fetal risk.
This index advertises many emergency, invasive, irreversible, and specialist-only obstetric/gynecologic procedures, including surgery, obstetric emergencies, medication protocols, and resuscitation topics, but it lacks a prominent warning that the material is for licensed clinicians and not for self-treatment or lay use. Because the collection is framed as directly usable guidance via natural-language prompting, an unqualified user could treat it as actionable medical instruction, creating substantial risk of bodily harm or delayed emergency care.
The skill recommends urgent clinical actions and names prescription-level treatments such as oxytocin and antibiotics without clearly directing the user to seek immediate evaluation by a qualified clinician rather than self-manage. In an obstetric postpartum context, this is dangerous because abnormal lochia, fever, uterine tenderness, or foul odor may indicate retained products of conception, postpartum hemorrhage, or puerperal infection, and delay or inappropriate self-treatment can rapidly worsen outcomes.
The skill provides concrete dosing instructions for prescription ovulation-inducing agents, hCG injections, and progesterone therapy without an explicit warning that these treatments require clinician supervision and should not be self-administered. In a medical skill about fertility and abnormal bleeding, omission of safety guardrails materially increases the risk of unsafe self-treatment, misdiagnosis, medication adverse effects, and delayed evaluation of other causes of infertility or bleeding.
This skill provides concrete diagnostic criteria and treatment direction for a fertility-related gynecologic condition, including ultrasound interpretation, progesterone thresholds, and a specific hCG dosing range, but it does not clearly warn that these steps require licensed clinical oversight. In this context, users could mistake the content for stand-alone medical instruction and act on it without appropriate evaluation, creating risk of misdiagnosis, inappropriate hormone use, and harm from missed alternative causes of infertility or contraindications.
This skill delivers actionable neonatal hepatitis B prevention guidance, including drug/biologic timing and dosing, without an explicit warning that it is medical information requiring clinician verification and local guideline review. In a neonatal care context, omissions or jurisdiction-specific differences can directly lead to mistimed HBIG/vaccine administration, missed prophylaxis, or inappropriate delays, creating patient safety risk.
The skill’s manifest and body describe a textbook editorial-team lookup function, while the declared skill metadata says it is a registry of 187 obstetrics and gynecology clinical skills. This mismatch can cause the agent to invoke the wrong skill, return irrelevant information, or bypass user expectations and higher-level routing controls that depend on accurate skill labeling.
This skill generates patient-specific chemotherapy regimens, including drug selection, dosing basis, treatment cycles, route of administration, and optional targeted therapy, but it does not include an explicit warning that the output is high-risk clinical guidance requiring oncology review. In this context, omission of a strong clinician-verification safeguard is dangerous because dosing or regimen errors in cytotoxic chemotherapy can directly cause serious patient harm, undertreatment, or death.
No suspicious patterns detected.
