Ni Haixia Classical Chinese Medicine Skill
Injects the clinical reasoning of master physician Ni Haixia into an AI Agent, enabling six-meridian pattern diagnosis and classical-formula prescribing from his perspective.
Purely local text knowledge base: no external calls, no dependencies, no destructive defaults—least privilege; footer disclaimer retained and states it does not replace licensed physicians. Deducted: style rules ban in-body medical safety disclaimers (confined to a footer box), which weakens risk signaling for a medical-advice skill; cites real dosages and potent drugs (aconite, ephedra) without user confirmation or escalation mechanisms; publisher unverified.
Clear retrieval flow (keyword index → Grep → Read → declare gaps), eight-point self-check, article-numbering disclaimers, honest marking of missing chapters—internally consistent. Deducted: all rules rely on model self-enforcement, no test suite or CI reproduction; the very complex format rules (eight checks, mandatory cards, row completeness) have unverified failure behavior on abnormal input; static cap 10.
Explicit trigger words, clear scenarios (differentiation, formula lookup, wellness Q&A) and retrieval paths; natively Chinese; declares honest gaps and no fabrication for missing material. Deducted: no explicit non-fit boundaries (emergencies, pediatrics, pregnancy), so it may be over-triggered for general medical queries.
SKILL.md is well layered (quick card → distilled highlights → navigation index → FAQ) with detailed changelog, version history, audit notes and path indices. Deducted: inconsistent figures across the repo (849 vs 1257 cases; 71 vs 72 Neijing chapters; 345 vs 374 herbs; mismatched version/platform statuses); license metadata unknown despite badge claiming MulanPSL-2.0; maintenance responsibility rests on an unverified personal repo.
Large, highly structured knowledge base with quick-reference tables, case tables and dosage comparisons offering real marginal value to target users. Deducted: usability depends on the model executing heavy style constraints; blind-test and 9-dimension review claims are self-reported and unverifiable; medical outputs necessarily need professional review; static cap 7.
Provides source line numbers, audit notes, dosage-correction records and numbering-scope statements, allowing partial traceability. Deducted: all evidence is self-produced/self-reviewed (blind tests, '100+ wins 0 losses' with no reproducible record); no third-party execution evidence; many inline citations cannot be checked in this static review; static cap 5.
- This skill outputs role-played TCM content including real dosages and potent drugs (raw aconite, ephedra); it must never replace licensed medical care. Safety notices are rule-moved to the footer—users must read the disclaimer.
- Multiple statistics are inconsistent across the repo (849 vs 1257 cases; 71 vs 72 Neijing chapters; 345 vs 374 herbs; conflicting version statuses). Verify figures against source files before citing.
- Blind-test and review results are self-reported and not independently reproducible; license metadata is unconfirmed—verify MulanPSL-2.0 applicability and copyright boundaries of the lecture materials before commercial use.
What does this skill do, and when should you use it?
This is a large domain-specific Agent Skill distilled from the teaching materials of Ni Haixia (1954-2012), a Taiwanese classical-formula (jingfang) physician. Its knowledge base covers all 129 lines of the Shanghan Lun, 23 chapters of the Jingui Yaolue, 72 chapters of the Huangdi Neijing, 374 herbs from the Shennong Bencao Jing, 1,257 structured medical cases plus 243 narrative cases, drawn from 2,452 pages of lecture notes. SKILL.md serves as the entry point, routing the model through keyword indexes to grep the modules/ and cases/ text corpus, then answer in Ni Haixia's first-person teaching voice. It enforces a six-meridian plus eight-principles diagnostic framework with detailed output rules for dosages, formula citation cards, and a mandatory disclaimer. It is a study-and-research tool, explicitly not medical advice.
The skill runs no scripts and makes no network calls. It works by: reading the quick-reference tables embedded in SKILL.md (six-meridian diagnostic formulas, six cold formulas, yuan-source point table, five shu-point formulas, etc.) and a keyword index; using grep/file reads across 14 modules (Shanghan Lun, Jingui, Neijing, acupuncture and materia medica, astrology/Yijing, Fuyang forum lectures, Zhongjing core methods, Stanford lecture) and a structured cases library to locate source passages; then producing answers under strict formatting rules — conversational Ni-style narrative as the body, at least one table or bolded key-point block, a standardized diagnosis line, a three-line citation card (source passage, original formula, clinical dosage) for every formula referenced, dual ancient/clinical dosage systems, and a fixed end-of-answer disclaimer. It requires the model to declare when content is missing from the corpus rather than fabricate, and forbids web search.
- Students of classical Chinese medicine who want Q&A study of Shanghan Lun and Jingui passages with Ni-style commentary and dosage comparisons
- Jingfang enthusiasts facing concrete symptoms (sweating vs. no-sweating colds, insomnia, constipation) who want a six-meridian diagnosis plus full formula composition and decoction method
- Practitioners-in-training searching real case material, e.g. querying the 1,257-case structured table by diagnosis or formula for cancer or cardiovascular cases
- Acupuncture learners needing quick lookup of yuan-source points, five shu points, the meridian-clock, and point indications
- Readers interested in Ni Haixia's signature positions: critique of the warm-disease school, yang-supporting theory, the 'seven cancers share one origin' doctrine
What are this skill's strengths and limitations?
- Exceptionally large corpus: 2,452 pages / 3.5M characters spanning classics, acupuncture, cases, and metaphysics, with structured indexes and grep guidance
- Very detailed output governance: dual dosage systems, formula-completeness ironclad rules, citation cards, anti-fabrication and no-web-search clauses that meaningfully curb hallucination
- Active iteration (v1.0.0 through v2.3.1) with documented errata (full dosage audit of 258 formulas) and author-run blind-test reports
- Multi-runtime distribution: ClawHub, SkillHub, Tencent IMA, and manual git install
- The SKILL.md entry file is very large (~107KB per the README) and the retrieval rules are elaborate — costly for small-context models
- License metadata conflict: GitHub shows 'unknown' while README badges state MulanPSL-2.0; verify before redistribution
- Second development of another open-source project (huoyalong/nihaisha-skill); content is roleplay/study material, unsuitable for actual clinical decisions
- Known data-quality caveats inside the corpus: some mislabeled case entries and missing Neijing chapters (ch. 25 and most of 66-74) — declared in the skill but users should stay alert
- No independent automated test suite; the README's blind-test results are self-reported
How do you install this skill?
Per the README: recommended via ClawHub — openclaw skills install @jangviktor-web/nihaixia or npx skills add https://clawhub.ai/jangviktor-web/skills/nihaixia; or manually — git clone https://github.com/jangviktor-web/nihaixia.git then copy the folder to ~/.claude/skills/nihaixia/. Tencent SkillHub and the mobile Tencent IMA store are also documented (V2.3.1 is the latest release). The entry point is SKILL.md; there are no script dependencies.
How do you use this skill?
Place the repo in your Agent Skills-compatible client's skills directory, then activate with trigger phrases such as: 'How would Ni Haixia view sweaty hands and feet?', 'Haixia perspective: I have a cold, chills, no sweating, stiff neck', or 'Analyze this insomnia case with jingfang thinking'. The model retrieves from the local corpus first, then answers in Ni's voice. Direct quick-reference questions also work ('When do I use Xiao Chaihu Tang?', 'Difference between raw and prepared aconite?').
How does this skill compare with similar options?
The README positions this alongside the same author's companion skills: Li Ke skill (acute/critical-care focus, aconite up to 750g), Hu Xishu skill (six meridians from eight principles, formula-pattern correspondence), and Wu Jutong skill (warm-disease triple-burner diagnosis) — complementary to Ni's general-practice teaching corpus. Related projects include tcm-distiller (a TCM reasoning-distillation pipeline) and an Android app (nihaixia-app) using the same knowledge base offline.