Every web application is a state machine. This is the foundational insight behind Boozang.
Users don't script click sequences. They model their application's behavior - defining modules, operations, and interdependencies. Tests become explorations of state space, not blind click sequences.
All documentation should reflect this paradigm. See .agent/context/testing-philosophy.md for details.
Documentation site for Boozang (Docusaurus).
npm run start # Dev server at localhost:3000
npm run build # Production build
npm run serve # Serve production build| Path | Purpose |
|---|---|
docs/ |
Main documentation (Markdown/MDX) |
static/agent/ |
Public AI agent docs (sync with bz repo) |
sidebars.js |
Navigation structure |
docusaurus.config.js |
Site configuration |
chatgpt/ |
LLM content processing |
| Command | Action |
|---|---|
start session |
Read sync file + instructions, check journal, remind context |
end session |
Commit, update sync + journal, log hours, summarize |
push |
Commit to current branch and push to origin |
pull |
Pull latest changes from origin |
journal |
Update work log and instructions |
scout <area> |
Analyze a codebase area and produce a structured report |
See .agent/instructions.md for comprehensive documentation.
User-facing AI documentation served from both sites:
- Docs site: https://docs.boozang.com/agent/AGENT.md (source:
static/agent/) - Main app: https://ai.boozang.com/agent/AGENT.md (source:
bz/public/agent/)
Content is mirrored from main bz repository. Keep in sync when updating.
docs/ - Main documentation (Markdown/MDX)
chatgpt/ - Processed content for ChatGPT/LLM training
pages/ - Cleaned markdown pages
static/ - Static assets (images, etc.)
agent/ - Public AI agent docs (mirrored from bz/public/agent/)
src/ - React components and CSS
blog/ - Blog posts
transcripts/ - Video transcripts
npm run start # Dev server at localhost:3000
npm run build # Production build to build/
npm run serve # Serve production build locally
npm run clear # Clear Docusaurus cache- Docs are in
docs/organized by topic (api-testing, core-concepts, etc.) - Use relative links between docs:
[link](../folder/file.md) - Images go in
docs/images/- reference as - Sidebar config is in
sidebars.js
The chatgpt/ folder contains scripts to crawl and clean documentation for LLM consumption:
crawler.js- Crawls the docs siteclean-en.mjs- Cleans and processes contentpages/- Output cleaned markdown files
- Required gh user:
ljunggren - Before any
git push, verify the active gh account:gh auth status 2>&1 | grep 'Active account: true' -B3 | head -1 - If the active account is not
ljunggren, rungh auth switch --user ljunggrenbefore pushing.
- L0 — Autonomous: Fixing typos, formatting, updating screenshots. Just do it.
- L1 — Proceed, tell me: Editing existing doc pages, improving explanations, updating examples. Do it, but explain what changed.
- L2 — Propose first: Creating new doc sections, restructuring navigation, changing the sidebar. Show the plan first.
- L3 — Full stop: Modifying Docusaurus config, changing deployment settings, altering the ChatGPT content pipeline. Stop and wait for approval.