bazi-skill

---
name: bazi-skill
description: Specialized workflow for BaZi 八字, Four Pillars 四柱, optional Zi Wei Dou Shu 紫微斗数 evidence, optional Western astrology / zodiac / 星座 evidence, optional NaYin 纳音, branch-relation 刑冲合害, Qi Men Dun Jia 奇门遁甲, Liu Yao 六爻 evidence, K-line style fortune JSON, true solar time 真太阳时, compatibility/synastry 合盘合婚, auspicious date/hour selection 择日择时, professional report writing, structured reports, 命理研报, and multi-school master/orchestrator workflows 多流派大师/主理规划师/主理官综合. Use when the user asks to build, debug, validate, or improve a host project that computes chart facts with code and asks AI only to interpret confirmed facts.
---

# bazi-skill

Use this skill for project-aware BaZi, optional Zi Wei Dou Shu evidence, optional Western astrology evidence, optional common-school evidence, and K-line style fortune workflows. Preserve the source-of-truth split: local libraries calculate charts; AI writes structured interpretation against a confirmed chart.

## Workflow

1. Classify the task:
   - Code change in a host project: read the repository docs and the touched modules before editing. Start with `CLAUDE.md` if present, then the relevant `components/CLAUDE.md`, `services/CLAUDE.md`, `types/CLAUDE.md`, or `constants/CLAUDE.md` if present.
   - Skill or workflow architecture change, including requests to reference `ai-berkshire`, a multi-agent research/team pattern, multi-school masters, 大师会诊, 主理规划师, or 主理官综合: update this skill or related skill docs first; do not change frontend/backend application code unless the user explicitly asks for implementation.
   - Analysis JSON generation, repair, or validation: read `references/project-contracts.md`, then run `scripts/validate_analysis_result.py` on any candidate JSON.
   - BaZi rules, prompt wording, scoring logic, or domain explanation: read `references/bazi-domain-reference.md`, then `references/analysis-methods.md` when the task needs interpretation or ranking.
   - 五行生克、十神、阴阳/大小（如小水、大火）、纳音、藏干、旺相休囚死、刑冲合害，或"今天/某天对我的命盘怎样": once the pillars are known, run `scripts/ganzhi_facts.py` and put its output into the evidence packet. Never derive these from memory or by reasoning over a table in context; quote the script output instead.
   - 盲派象法, 象法体系, concrete imagery, 宫位象, 十神象, or requests to make 象法 more complete/source-bounded: read `references/xiangfa-system/source-map.md`, then `references/xiangfa-system/coverage-map.md`, then only the required slice files.
   - True solar time, strict apparent solar time, timezone, longitude correction, equation of time, or boundary-hour issues: read `references/true-solar-time.md`; if implementing, also read `references/project-contracts.md`.
   - Zi Wei chart logic, palace display, or pattern wording: read `references/ziwei-reference.md`.
   - Western astrology, zodiac signs, star signs, 星座, 西洋占星, natal chart, ascendant, moon sign, aspects, transits, or astrology synastry: read `references/western-astrology.md`.
   - NaYin 纳音, branch relations 刑冲合害, Qi Men 奇门遁甲, Liu Yao 六爻, Mei Hua 梅花易数, Da Liu Ren 大六壬, modern astrology, traditional astrology, or other school extensions: read `references/common-schools.md`; then load only the matching prompt/reference files.
   - Compatibility or synastry analysis, 合盘, 合婚, 伴侣匹配, relationship fit, or partnership matching: read `references/compatibility-analysis.md`, plus `references/analysis-methods.md`; if the task is also app implementation, read `references/project-contracts.md`.
   - Auspicious date/hour selection, 吉日吉时, 择日, or 择时: read `references/auspicious-timing.md` and `references/analysis-methods.md`; if the task is also app implementation, read `references/project-contracts.md`.
   - Professional report, 命理研报, structured report, Markdown report, or HTML report workflow: read `references/report-generation.md` and `references/project-contracts.md`; if using an `AnalysisResult`, validate the JSON before composing the report.
   - 古籍条文、典籍引用、条文出处、引用核对，或需要为判断补依据: read
     `references/classics/index.md`, then only the topic card files your school needs;
     verify with `scripts/validate_citations.py`; locate raw source text with
     `scripts/search_classics.py --corpus`. Never read `references/classics/corpus/`
     end to end.

2. Apply an information-completeness gate before analysis, JSON generation, report rendering, or multi-agent work:
   - If required user information is missing, ask follow-up questions before proceeding. Do not guess birth facts, chart facts, event constraints, relationship counterpart data, or report scope.
   - Ask only the missing essentials, preferably 1-3 concise questions at a time.
   - For normal BaZi calculation, required essentials are: gender, birth date, birth time, calendar type (solar/lunar if ambiguous), birthplace plus longitude/timezone basis, and whether true solar time should use the current legacy method or strict apparent solar time when that distinction matters. City-level birthplace is acceptable by default only when it resolves to a longitude; for boundary-hour or professional requests, prefer district/township, map point, hospital, or manual longitude. For Zi Wei, first verify that the current repo actually contains a tracked implementation and contract; otherwise treat Zi Wei as unavailable or planned. For Western astrology, first verify a tracked ephemeris/service or require a user-confirmed astrology chart; otherwise treat it as unavailable or planned.
   - For professional-mode chart input, required essentials are: four pillars and gender. Birth year, startAge, direction, and daYun can use documented defaults only when the user accepts approximate/professional-mode behavior.
   - For compatibility/synastry, require both people’s chart inputs or confirmed charts plus the relationship goal/context.
   - For auspicious timing, require event type, candidate date range, location/timezone, and any hard constraints before ranking dates/hours.
   - For Qi Men or Liu Yao, require a computed/user-confirmed plate or hexagram plus the exact event/question; do not cast or derive the plate/hexagram in AI text.
   - For professional reports, require the source `BaZiResult`, `AnalysisResult`, or report-spec JSON and the desired output format/scope before rendering.
   - Optional missing details may be handled with explicit assumptions only when the final output remains contract-valid and the user request is not asking for precision.

## Multi-School Masters + Orchestrator Pattern

Use this pattern when the user asks for an `ai-berkshire`-style team workflow, multi-agent analysis, 多流派大师, 大师会诊, 主理规划师, 主理官综合, deep review, report generation, or a complex task where different mainstream schools should be compared. For role details, read `references/agent-roles.md`; for executable school prompts, read `references/school-prompts/index.md` and only the selected role prompt files.