Security audit
Security checks for vulnerabilities and agentic risk
The skill promises not to save sensitive birth details but then requires saving them in identifiable local reports.
Before supplying birth details, require archiving to be optional and explicitly approved, with a chosen storage location and deletion process. Setup downloads npm dependencies. Calculation defects also mean the readings should not guide consequential decisions.
SKILL.mdMandatory archiving exceeds the disclosed data-retention consent
Birth information remains in a local archive beyond the disclosed processing purpose. No external transmission or access by another party was established.
The code is within the declared divination domain, but its central calculation materially differs from the promised classical-grounded engine. It selects a hexagram using only moving-line and yang-line counts; for example, six static yang lines yield 师 rather than 乾. It also hard-codes the day stem, assigns line branches sequentially, and defaults most self/responding relationships. Its element lookup does not recognize heavenly stems, so the day-stem argument silently defaults to earth. These are substantive departures in the core advertised calculation, not merely supporting implementation details. Other advertised engines cannot be assessed from this chunk, and their absence here is not treated as a mismatch.
The instructions authorize npm installation at setup and on missing-module errors despite describing a no-network, bridge-only execution boundary. Installation can contact external registries and execute dependency lifecycle scripts, expanding execution beyond the stated local scripts; this does not establish that any dependency is malicious or that birth data is transmitted.
Mandatory dossier creation directly contradicts the collection-time promise that sensitive birth information will not be persistently stored. The prescribed filename itself reveals birth date, birth-time category, and sex, while the dossier can retain additional personal information accessible through shared storage, backups, or synchronization.
This is the same privacy vulnerability as SDI-4, not an additional independent exposure. Users consent to local calculation under an explicit no-persistence assurance, but the workflow subsequently mandates a durable archive containing sensitive birth information.
This duplicates the mandatory-archiving issue identified by SDI-4 and SQP-2. Even without network transmission, unexpected retention can expose personal information to other local users, indexing services, synchronized folders, or backups.
The engine selects a hexagram using only the number of moving lines and yang lines, discarding their positions; distinct six-line configurations therefore produce the same hexagram, and moving status incorrectly influences the primary hexagram. This undermines output integrity in a skill advertised as computing charts grounded in classical rules, potentially causing the agent to present incorrect readings as authoritative calculations. The explicit simplification comments suggest negligence rather than malicious intent; the impact is misinformation, not demonstrated code execution or data exposure.
The comment states that 天禽 always occupies the central fifth palace, but the implementation assigns nineStars[4], which is 天英 (L016); 天禽 is nineStars[8] (L020). This directly contradicts the documented chart-construction rule and propagates into the reported palace placement and auspiciousness assessment.
The manifest promises classical-source-grounded divination, but this implementation determines yin/yang dun using fixed calendar cutoffs and selects the ruling star and door by cycling elapsed hours modulo nine and eight from an arbitrary epoch. It does not compute the Qi Men chart configuration or xun needed to derive these values, yet presents them as actual 奇门遁甲 results (L459–L462) and uses them to rank recommended dates (L322–L335), without identifying them as approximations.
analyzePlate passes the twelve Zi Wei palaces to calculateStrengthEnhanced and calculateYongshenEnhanced, whose BaZi calculations count stems, branch roots, and hidden stems as though these were the four birth pillars. This produces systematically unreliable strength scores and useful-god recommendations presented as classical-source-grounded analysis, making it an output-integrity defect rather than a code-execution or data-exfiltration vulnerability. The skill's emphasis on authoritative chart interpretation increases the risk of misplaced trust, although its exclusion of medical, legal, and investment recommendations limits the intended impact.
Overly Broad Trigger: description clause 'Use whenever the user asks to 算命 / 算卦 / 批八字 / 看生辰八字 / 排盘 / 看命盘, asks about 大运 / 流年 (luck cycles) or about 运势 only in a chart context' activates on 'to', which is too short and may match unintended inputs
Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.
"private": true,
"type": "module",
"dependencies": {
"cantian-tymext": "^0.0.26"
}
}
The scoring disclosure at L388–L389 says that only day-master and year-branch relationships contribute and explicitly excludes heavenly-stem combinations (天干五合). However, calculateOverallScore adjusts the score by +10, −5, or −10 for those combinations at L307–L309, contradicting the report's stated scoring basis.
Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.
"bazi": "node bazi-analysis.js"
},
"dependencies": {
"iztro": "^2.5.8"
}
}
Detected: suspicious.dangerous_exec