Compression King
iMandalArt 2.2
Turn any source into one five-character center and eight orthogonal angles. Preserve の字型 identity for downstream indexes and Routes.
source → center + eight angles
Open iMandalArt
BIRD · FIRE · MANDALA · PAPER
TWHSI
Agent
Skills
Let AI handle structure.
Keep human judgment slow.
Three Kings 2.2: compress with iMandalArt, address with BIRD, and publish with an A4 eight-page booklet.
registry: twhsi/skills
revision: loading
generated: loading
skills: loading
TIMECARDSLLMDESKTOPPUBLISH
Permanent notes become a book.
Book Address · W+T+K+A Index · Route · Exact Deep Link · Semantic Role
Skills 三大天王 2.2
One continuous knowledge-publishing route: compress the material, give every idea an address, then return it to a verified paper object.
Open skill-tree.jsonAddress King
BIRD Book Deconstructor 2.2
Assign Book Address, W+T+K+A Knowledge Index, Route, exact Deep Link, and optional Semantic Role. Export to TheBrain, Excel, Roam, or paper.
angle → address → Route
Open BIRD
Publishing King
A4 Eight Page Booklet 2.2
Place the overview on page 1 and seven BIRD Routes on pages 2–8. Deliver reading PDF, print sheet, editable DOCX, previews, and validation.
overview + 7 Routes → paper book
Open A4 Booklet
2.2 New
Shared handoff, stable formats
Center identity, spatial order, source IDs, citations, addresses, and Routes survive the whole chain. BIRD-2.1 and BookletManifest 2.0 remain compatible.
compress → address → publish → validate
Copyable Demo
A compact workflow shows the repository's purpose: AI handles structure, humans keep judgment.
GitHubOne line promise
AI Agent Skills for Chinese Knowledge Workers
Turn daily plans, weekly reviews, manuscript notes, and Markdown drafts into agent-ready workflows: BIRD 2.1 addresses, concise key points, iMandalArt cards, FIRE analysis, A4 booklets, and publishing paths.
Input: manuscript + permanent notes + PDF Skills: imandalart 2.2 + thebrain-bird-address 2.2 + a4-eight-page-booklet 2.2 Output: eight angles + BIRD Routes + verified A4 mini book
Start from the compact context file, then route to the right skill.
curl -s https://www.twhsi.com/llms.txt
Five seconds should reveal the project category, flagship skill, and proof of workflow.
iMandalArt + FIRE + planning + publish
Visual proof will come after the message is stable: screenshots, share cards, and example outputs.
no screenshots in phase one
Keyword Graph View
Paste an article, extract exactly eight context-sensitive keywords, and inspect a centerless weighted network with definitions and evidence.
Open live tool
Live Skill
Eight keywords. No center. One weighted network.
Blacklist structural terms, generate the graph, then click any node to read its definition, note, evidence sentences, and strongest connections. Edge color and thickness both show weight.
Traditional Chinese, English, and mixed text
Rainbow weights from purple W1 to red W9
Zoom controls and fixed full-screen workspace
Weekly Reverse Review
A nearly twenty-year weekly review practice converted into an AI-assisted Skill: read YEAR, Week, Day, and Inbox evidence, then build one quiet and vivid 8 Big Rocks plan.
GitHubNew skill
Let AI spend tokens so the human keeps quiet judgment.
Weekly Reverse Review gathers annual plans, hundred-year life plans, last week's plan and review, seven days of daily plans and diary notes, calendar evidence, and inbox noise. The goal is not to add more tasks. The goal is to ask what brings real happiness and peace.
Lower cognitive load: an 11-step review becomes lighter.
Four evidence angles: YEAR, Week, Day, and Inbox are compared first.
Reverse wish list: shrink, slow down, delete, defer, or keep presence.
Default answer: speak less, stay present, move slowly.
ⒻInner Release ⒸMoney Noise ⒼLearning Depth Speak less Review first Number lines Prove nothing Fund the work Build index ⒷManuscript ◎◎◎◎◎ ⒹFamily Spark Pull one center Peaceful Week Ask not lecture Draft before all ◎◎◎◎◎ Walk the long ⒺPeople Path ⒶHealth Reset ⒽJoyful Rest Keep three cards Move with care Sing and loosen Connect next Build rhythm Guitar and sun
iMandalArt 2.2
The featured skill in this registry: a hard-line 3x3 thinking card that lets mainstream LLMs turn loose notes, plans, and source material into one compact CJK-friendly Mandala index.
GitHubFeatured skill
One center, eight orthogonal angles, eleven physical lines.
iMandalArt 2.2 is optimized for CJK workflows: each surrounding content phrase is exactly five Han characters, each title is compact, and full-width separators keep the card stable in chat previews, TheBrain/Cerebro notes, Hermes, Discord, and clipboard-based workflows.
The format is currently CJK-friendly. An English-native version is planned so the same 3x3 thinking rhythm can work naturally for English notes without forcing a Chinese character contract.
Weekly planning
Writing focus
Knowledge capture
TheBrain notes
Ⓐ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
A4 Eight Page Booklet 2.2
One A4 sheet becomes eight upright reading pages, one folded paper object, and one route from structured knowledge back to the hand.
GitHub
Paper publishing skill
One sheet. Eight pages. One thinking route.
Compile text, BIRD Graph JSON, iMandalArt, FIRE notes, daily plans, outlines, images, templates, or existing PDFs into a reproducible paper booklet.
Permanent Note→BIRD Address→BookletManifest→Folded Booklet
8 upright page PDFs
8-page reading PDF
A4 landscape 6-7-8-1 / 5-4-3-2 imposition
Editable DOCX and visual validation
Open the Skill
Latest Skill Updates
Every card is generated from Git history and skill metadata: semantic version when declared, latest commit revision, and last update time.
/skills.jsonLLM Endpoints
These are the canonical machine-readable entry points for another LLM agent or automation client to continue the work without reading the whole repository first.
/agent.json| Endpoint | Use | Type | Copy |
|---|---|---|---|
| /agent.json | Canonical manifest with registry identity, routes, skill metadata, and LLM agent contract. | application/json |
|
| /skills.json | Generated skill index from skills/*/SKILL.md, including versions and update timestamps. |
application/json |
|
| /llms.txt | Compact LLM context with routes, skills, version labels, update times, and GitHub links. | text/plain |
|
| /install | Human and LLM install guide for local runtime usage and static website deployment. | text/html |
Skill Routes
Routes help LLM agents select the right skill and help humans understand how the operating system is organized.
View allTime/time
Daily focus, weekly review, planning rhythm, calendar actions, and long-range training loops.
weekly-reverse-review todays-daily-plan imandalart personal-athlete-81-grid
Cards/cards
FIRE analysis, grid cards, Markdown tables, and graph views.
weekly-reverse-review fire-analysis-card markdown-nine-grid-clipboard obsidian-graph-view
System Map
The map is the human-readable layer: it shows where skills sit inside a broader writing, planning, desktop, and publishing workflow.
GitHubHermes is the cockpit, not a toolbox.
The repository keeps the source files, the website publishes live metadata, and the map helps a human decide which work should become a reusable skill, a book note, a card, or a desktop routine.
Skill Index
Search the generated registry. Each row includes semantic version, latest Git revision, update time, route axes, description, and install command.
loading| Skill | Version | Updated | Axes | Description | Install |
|---|---|---|---|---|---|
| fire-analysis-card | v2.0 |
loading | Cards | Prepare card-box material for semantic search. | cp -R skills/fire-analysis-card ~/.codex/skills/ |
Install & Deploy
Use GitHub as the source of truth, Vercel as the static build host, and the generated JSON files as the LLM entry points.
/installGitHub
Keep every skill in versioned Markdown and let the build step derive registry metadata from the repository.
twhsi/skillsVercel
Build with npm run build and serve dist. The live site publishes JSON, text, and HTML surfaces.
Custom Domain
Point www.twhsi.com to the Vercel project so the registry stays stable for humans and LLM agents.
$ registry live
versions generated
timestamps published
multi-LLM