nihaisha: Turn Ni Haixia’s TCM Courses into a Searchable Agent Skill for AI

The core question this article answers: How can you enable AI assistants like Claude Code or Codex to systematically retrieve all of Ni Haixia’s Traditional Chinese Medicine (TCM) course materials – and help you with pattern differentiation, formula comparisons, and screenshot evidence – using natural language, while keeping everything traceable and safe?

For TCM learners and technical engineers, Ni Haixia’s curriculum is massive – from Shanghan Lun and Jingui Yaolüe to Huangdi Neijing, Shennong Bencao, acupuncture, Tianji, clinical cases, and the Fuyang Forum. That’s over a dozen modules and thousands of board screenshots. Traditional document search is inefficient, and it’s hard for AI to directly “understand” the internal terminology and diagnostic logic. The nihaisha Agent Skill solves this by distilling the courses into a structured, transferable, and actionable format. It lets you use plain language to perform professional TCM course retrieval and study organization.

Course Distillation Method: From Raw Lectures to an Agent Skill

The question this section answers: How do you transform high‑density course materials into an executable, traceable AI skill?

nihaisha uses the distillation method from the author’s lineage-skill project. The core idea is to unify heterogeneous materials – raw lecture transcripts, video screenshots, notes, e‑books – into Markdown indices and evidence chains that an AI Agent can directly read. Each course module has a main file under references/ (e.g., shanghanlun.md, jingui.md) and a separate screenshot evidence file (e.g., shanghanlun-screenshot-evidence.md). The screenshot evidence records the course module, lesson number, timestamp, formula name, acupoint, or pathogenesis keyword for each image, so the AI can quickly locate them via scripts or semantic matching.

The advantages of this method: traceability (every piece of knowledge points back to a specific lesson and screenshot), transferability (the Skill directory can be installed on different Agents like Codex, Claude Code, or OpenClaw), and actionability (the AI can generate study plans, formula comparison tables, acupoint summaries, and more directly from these materials).

Course Modules and Coverage

The question this section answers: Which of Ni Haixia’s courses does nihaisha actually include, and how complete is the data?

As of now, nihaisha has fully integrated the following course modules with both text and screenshot evidence:

In addition, it includes text transcripts of the Liang Dong interview and Stanford University speech (screenshots to be added later), plus several written note modules: Acupuncture Great Compendium Notes, Huangdi Neijing Notes, Shennong Bencao Notes, Shanghan Lun Notes, Jingui Yaolüe Notes, as well as Hantang Chinese Medicine articles, Diagnostic Logs, E‑book Collection, and Audio Collection. In total, there are 2,986 screenshot evidence entries, covering most of the whiteboard drawings, practical demonstration images, and key formula‑pattern texts from the courses.

Core Capabilities: Professional Retrieval with Natural Language

The question this section answers: What exactly can you do with this Skill? Can you give concrete examples from everyday study?

1. Plain‑Language Entry Point

You don’t need to know terms like “Taiyang中风” (wind stroke) or “dan yu mei” (desire to sleep) first. You can simply ask: “I have a cold, feel chilly, have no sweat, and my body aches – what is that?” The Agent maps this to “Taiyang cold damage” and the Mahuang Tang pattern. Or ask “cold hands and feet, diarrhea, can’t sleep” – the Skill guides you into the Shaoyin or Taiyin differentiation framework. This translation layer greatly lowers the barrier for beginners.

2. Six‑Channel and Formula‑Pattern Navigation

shanghanlun.md organizes the core content of Shanghan Lun by the six channels, symptoms, formulas, and disease progression. For example, you want to compare the pattern differentiation of Guizhi Tang, Mahuang Tang, and Gegen Tang. The Agent can return a symptom‑cluster comparison table, the pathogenetic mechanism hierarchy, and contraindication reminders. This is extremely efficient for review or note‑taking.

3. Acupoint and Herbal Studies

In the acupuncture module, you can ask: “What are the common emergency acupoints on the Ren and Du meridians?” The Agent will search acupuncture.md and related screenshot evidence, returning the acupoint location, needling depth, main indications, and Ni’s combination ideas. In the Shennong Bencao module, you can ask: “What are the properties, processing method, and compatibility taboos of Fu Zi (Aconite)?” The Skill returns single‑herb information, dosage forms, and safety warnings emphasized in the lectures.

4. Lesson‑by‑Lesson Review and Topic Maps

If you want to review lesson by lesson, the Skill can generate topic maps, keyword lists, and self‑test questions based on a course module (e.g., Jingui Yaolüe lessons 1–10). For instance: “Use nihaisha to organize the course threads on chest bi (painful obstruction), water qi, and phlegm‑fluid in Jingui.” The result includes each lesson’s core pathogenesis, formulas, and Ni’s clinical insights.

5. Screenshot Evidence Retrieval

This is a unique feature of nihaisha. Many TCM learners struggle to understand needling techniques, tongue diagnosis images, or formula whiteboards using text alone. The Skill includes a script scripts/search_screenshots.py that can search the screenshot index by formula name, acupoint, lesson number, pathogenesis keyword, or time stamp. Examples:

python3 scripts/search_screenshots.py Xiao Chai Hu Tang modification
python3 scripts/search_screenshots.py Shaoyin diarrhea
python3 scripts/search_screenshots.py Tianji life palace
python3 scripts/search_screenshots.py acupuncture Zusanli

The output gives relative paths inside assets/screenshots/ – the Agent can display them directly or further analyze the text on the images. All major modules (acupuncture, Shanghan, Jingui, Tianji, Huangdi, Bencao) already have screenshot evidence.

Installation: Adding the Skill to Your AI Assistant

The question this section answers: How do you install nihaisha on Claude Code or Codex?

Method 1: Let the AI Install It (Recommended)

Simply tell your AI assistant:

Help me install the nihaisha skill:
https://github.com/JuneYaooo/nihaisha-tcm

The Agent will clone the repository and install it into the appropriate skills directory.

Method 2: Manual Installation

git clone git@github.com:JuneYaooo/nihaisha-tcm.git
cd nihaisha-tcm
bash install_as_skill.sh --target codex    # For Codex
# or
bash install_as_skill.sh --target claude   # For Claude Code

The script installs the skill to:

• Codex: ~/.codex/skills/nihaisha/
• Claude Code: ~/.claude/skills/nihaisha/
• OpenClaw: ~/skills/nihaisha/

After installation, restart your Agent to reload the skill metadata.

Practical Examples: Conversations and Scripts

The question this section answers: Can you show complete example prompts and what the output looks like?

Here are prompts you can directly copy to your AI assistant (after installing nihaisha):

Use nihaisha to help me organize the differences between Taiyang wind stroke and Taiyang cold damage.

Use nihaisha to retrieve the pattern differentiation thresholds for Guizhi Tang, Mahuang Tang, and Gegen Tang.

Use nihaisha to explain in plain language: why do some people with a cold feel chilly and have no sweat, while others fear wind and have sweat?

Use nihaisha to find whiteboard screenshot evidence related to Xiao Chai Hu Tang.

Use nihaisha to find the course structure for chest bi, water qi, and phlegm‑fluid in Jingui.

Use nihaisha to organize the Ren and Du meridians and common emergency acupoints from the acupuncture course.

Use nihaisha to find whiteboard evidence for the life palace and four transformations in Tianji.

You can also run the screenshot search script directly (inside the skill directory):

python3 scripts/search_screenshots.py Xiao Chai Hu Tang modification
python3 scripts/search_screenshots.py Shaoyin diarrhea

Safety Boundaries: For Study Only, Not Medical Diagnosis

The question this section answers: Can I use this Skill to prescribe for myself? What are the risks?

nihaisha explicitly states: It is intended for course learning and TCM theory organization only, not for personal diagnosis, prescription, or dosage advice. When dealing with aconite‑containing formulas, Sini Tang, Dachengqi Tang (urgent purging to preserve yin), Didang Tang, Daxianxiong Tang, cancer/tumors, pregnancy, children, chest pain, altered mental status, severe dehydration, or other acute critical conditions, you must immediately consult a qualified physician or emergency services.

This safety boundary is crucial. The Skill provides the formula‑pattern relationships, pathogenetic analysis, and historical cases from Ni Haixia’s courses, but that is never equivalent to medical advice. Users should treat it as a “study companion,” not an “electronic healer.” The project documentation reinforces this repeatedly, and warnings appear throughout the installation and usage examples.

Reflection and Insights: Why Screenshot Evidence is the Soul of This Learning Tool

While distilling these TCM courses, I learned a hard lesson: text can never replace the information density of whiteboard drawings and practical demonstration images. Ni Haixia’s lectures have countless hand‑drawn meridian diagrams, tongue photos, needling angles, and formula derivations. Converting them to text loses much of the visual logic and spatial relationships. nihaisha chose to use a screenshot evidence index, tagging every image with its course module, timestamp, and keywords, so the AI can “see” and cite the image. This approach has a high upfront cost (nearly 3,000 screenshots), but the value to learners is enormous – you can directly see Ni’s reasoning on the blackboard, not a second‑hand paraphrase.

Another insight: An Agent Skill’s boundaries must be stricter than ordinary documents. Because AI actively generates suggestions, if the skill does not explicitly forbid diagnostic behavior, users may mistakenly believe the AI has medical competence. nihaisha embeds safety statements in every key location and even adds “contraindication reminders” in formula‑pattern comparisons. This design philosophy should be adopted by all medical‑related Skills.

Suitability at a Glance

Practical Summary & Action Checklist

One‑page summary:

• nihaisha is an AI Agent Skill for Claude Code, Codex, and OpenClaw.
• Contains text indices of 11 major Ni Haixia course modules and 2,986 screenshot evidence entries.
• Retrieve formulas, acupoints, herbal properties, pathogenesis, lesson numbers, and screenshots via natural language.
• Install with git clone + bash install_as_skill.sh --target codex/claude.
• Example usage: “Use nihaisha to compare Guizhi Tang and Mahuang Tang.”
• Scripted screenshot search: python3 scripts/search_screenshots.py keyword
• Strict safety boundary: no medical diagnosis; for acute conditions, see a doctor.

Action checklist (for new users):

1. Install the Skill (let your AI do it for you).
2. Restart your Agent.
3. Ask your first question: “Use nihaisha to list all formulas in Shanghan Lun.”
4. Try a plain‑language query: “I have a cold, chills, no sweat, body aches – what formula should I look up?”
5. Run the screenshot search script once to confirm it returns image paths.
6. Pick a module you are studying (e.g., Jingui) and ask: “Use nihaisha to organize the course thread on chest bi.”

Frequently Asked Questions (FAQ)

Q1: Which AI Agents does nihaisha support?
It supports Claude Code, Codex, and OpenClaw. Use the install_as_skill.sh script with the target flag.

Q2: Do I have to download all screenshot images separately?
No. The repository already contains compressed WebP images under assets/screenshots/. They are local after cloning.

Q3: Can I use it to prescribe for myself?
No. The Skill is for course learning and theory organization only, not personal diagnosis, prescription, or dosage advice. For urgent medical conditions, always consult a doctor.

Q4: Does it work offline?
Yes. All materials and screenshot indices are local; the Agent does not need an internet connection to search them.

Q5: Are there screenshots for the Liang Dong interview and Stanford talk?
Currently only text transcripts exist for those two modules. Screenshot evidence will be added later.

Q6: How do I update to the latest course materials?
Go into the skill directory and run git pull. If screenshot indices are updated, re‑run the search script.

Q7: Can I add my own notes or screenshots?
Yes. Append content following the Markdown format under references/ and update the corresponding screenshot evidence file. We recommend forking the repository for your own maintenance.

Q8: What is the relationship with the Datawhale or LINUX DO communities?
The project acknowledges those communities for their support of open‑source learning and knowledge sharing, but nihaisha is an independent project for learning and exchange only.