AI Agent Skills for Chinese Knowledge Workers

Reusable agent skills for keyword graphs, weekly review, iMandalArt, FIRE analysis, card-based notes, writing, publishing, and Chinese knowledge workflows.

iMandalArt, FIRE semantic analysis, planning, and publishing workflows for Claude Code, Codex, and mainstream LLM agents.

This repository is TWHSI's public agent-skill system for Chinese knowledge workers. It turns repeatable writing, planning, note-making, and publishing workflows into skills that can be used by Claude Code, Codex, ChatGPT, Gemini, Hermes, and other LLM agents.

The source of truth lives in skills/*/SKILL.md; the website turns those files into generated LLM manifests, a searchable skill index, install commands, version labels, Git revisions, and update timestamps.

What This Repository Is

twhsi/skills is an installable, shareable, LLM-readable skill registry. It turns long-running workflows for weekly review, iMandalArt, FIRE analysis, card-based notes, EPUB publishing, HyperCard returns, and desktop work into public procedures that agents can repeat and humans can inspect.

The point is not to make humans behave like faster machines. The point is to let AI spend tokens on repetitive structure, while humans keep judgment, relationships, pace, and the quieter question:

Which action brings real happiness and peace?

The newest featured skill is keyword-graph-view: it extracts eight context-sensitive keywords from an article and turns them into a centerless weighted network with blacklist filtering, node notes, and evidence.

Live site: https://www.twhsi.com/
Live Keyword Graph: https://keyword-graph-view.twhsi.chatgpt.site/

Agent manifest: https://www.twhsi.com/agent.json
Skill index: https://www.twhsi.com/skills.json
LLM context: https://www.twhsi.com/llms.txt

What It Does

• Convert long Markdown notes into iMandalArt 2.01 hard-line 9-grid cards.
• Apply FIRE semantic analysis for card-box thinking, retrieval, and writing structure.
• Turn permanent notes into BIRD 2.0 Book Addresses, Knowledge Indexes, Routes, Deep Links, and TheBrain manuscript scaffolds.
• Generate agent-readable metadata through agent.json, skills.json, and llms.txt.
• Route daily planning, weekly planning, booklet, EPUB, Markdown, and knowledge-management workflows.
• Extract eight keywords and render a distributed weighted graph with definitions, evidence, and blacklist control.
• Package Chinese-first workflows so LLM agents can operate on them consistently.

Copyable Demo

Input: 200 days of daily plans, manuscript notes, or Markdown drafts
Skills: imandalart + fire-analysis-card + project-note-json-to-epub
Output: FIRE JSON + iMandalArt 2.01 card + Markdown / EPUB / booklet path

Featured: Keyword Graph View

keyword-graph-view turns Chinese, English, or mixed text into a distributed network of exactly eight keywords. It removes blacklisted terms, weights co-occurrence edges from W1 to W9, and gives every node a definition, note, and source evidence.

Open the live tool: keyword-graph-view.twhsi.chatgpt.site

Featured: Weekly Reverse Review

weekly-reverse-review turns an 11-step weekly review into an AI-assisted reverse planning loop.

The skill reads four angles before writing the next week:

• YEAR: annual plan and hundred-year life plan.
• Week: last weekly plan and weekly review.
• Day: daily plans, seven-day diary notes, and calendar evidence.
• Inbox: loose tasks, reminders, subscriptions, errands, and collected noise.

It then asks what should be smaller, slower, deleted, delayed, or kept as presence instead of achievement. The center question is:

Which action brings real happiness and peace?

Default answer:

Speak less. Stay present. Move slowly.

For an LLM or agent, start with:

curl -s https://www.twhsi.com/llms.txt

Featured: iMandalArt 2.01

imandalart turns loose source material into one hard-line 3x3 Mandala index card.

iMandalArt 2.01 is designed around a strict text contract:

• Eight orthogonal surrounding angles labeled Ⓐ through Ⓗ.
• A double center axis displayed as ◎◎◎◎◎.
• Exactly 11 physical text lines, so the card survives chat previews, note apps, and clipboard workflows.
• Compact CJK-friendly cells for TheBrain/Cerebro, Hermes, Discord, Codex, and other LLM chat surfaces.

Current status: iMandalArt 2.01 is optimized for CJK workflows. An English-native version is planned so the same 3x3 thinking rhythm can work naturally for English notes without forcing a Chinese character contract.

Conceptual example:

ⒶHealth Reset  ⒷManuscript  ⒸMoney Noise
Move with care  Pull one center  Review first
Build rhythm    Draft before all Fund the work

ⒹFamily Spark  ◎◎◎◎◎    ⒺPeople Path
Ask not lecture Weekly Review Keep three cards
Walk the long   ◎◎◎◎◎    Connect next

ⒻInner Release ⒼLearning    ⒽJoyful Rest
Speak less      Number lines  Sing and loosen
Prove nothing   Build index   Guitar and sun

Use it for weekly planning, writing focus, knowledge capture, and CJK note workflows where visual stability matters as much as semantic compression.

Featured: Eight Page Booklet 2.0

eight-page-booklet turns text, notes, images, BIRD Graph JSON, iMandalArt grids, 64+8+1 structures, FIRE material, daily plans, Bike or Bonsai outlines, templates, and PDFs into a printable, cuttable, and foldable eight-page booklet made from one A4 sheet.

The skill does more than place content into eight panels. It separates content from layout, normalizes the source into BookletManifest 2.0, and generates every reading, printing, preview, and validation artifact from the same manifest.

• Page 1 can hold the cover and overview. In BIRD or 64+8+1 mode, it contains one directory nine-grid and one keyword nine-grid.
• Pages 2–8 can form seven Routes. Each page keeps one central claim, three to five concise points, and one ordered reading path.
• iMandalArt spiral-order coordinates are preserved together with source IDs, BIRD Addresses, citations, and cross-route notes.
• Outputs include eight upright single-page PDFs, an eight-page reading PDF, an A4 landscape print imposition, preview images, and a validation report.
• The fixed imposition order is 4, 3, 2, 1 upside down on the top row and 5, 6, 7, 8 upright on the bottom row.
• Four validation levels check structure, typography, imposition, and visual quality, including page numbers, Chinese-capable fonts, clipping, rotation, blank pages, and text overlap.

Full skill specification and installation source: skills/eight-page-booklet/SKILL.md

What This Is

• A GitHub-backed skill registry for Chinese agent workflows.
• A machine-readable index for LLM agents that need routing, install commands, and resource discovery.
• A human-readable map for deciding which workflows should become reusable skills.
• A static website that publishes each skill's semantic version when declared, latest Git revision, and latest update time.

Core Skill Stack

Current Metadata Highlight

fire-analysis-card is now on FIRE 2.0:

• F = Full-D: stable numbering for temporary, permanent, and project notes.
• I = Index: keyword webs and retrieval handles.
• R = Route: thinking paths through the material.
• E = Evolution: time-based card-box growth for semantic search.

Skill Versions And Updates

The live website generates freshness metadata from skill files and Git history on every deployment. Each skill entry includes:

• version: the semantic skill version when declared, otherwise Unversioned
• revision and revision_short: the latest Git commit for that skill path
• updated_at: the latest update timestamp for that skill path

Open the live generated table:

• https://www.twhsi.com/#updates
• https://www.twhsi.com/skills.json

Featured metadata check: imandalart declares v2.01, and fire-analysis-card declares v2.0.

thebrain-bird-address implements BIRD 2.