# Saizeriya CLI > Operate the saizeriya.js command-line interface on behalf of a user. Use when the user wants an AI agent to start or resume a Saizeriya mobile-ordering session, inspect session state, look up item codes, manage a cart, view account or receipt details, or call Saizeriya CLI commands with npx/bunx. This skill is for user-facing CLI operation, not package development, publishing, or source-code maintenance. - Skill: `pnsk-lab/saizeriya-cli` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add pnsk-lab/saizeriya-cli` - Raw SKILL.md: https://api.skillmd.com/api/skills/pnsk-lab/saizeriya-cli/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: pnsk-lab (https://skillmd.com/u/pnsk-lab) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/pnsk-lab/saizeriya-cli --- # Saizeriya CLI ## Operating stance Act as the user's careful CLI operator. Convert the user's dining/session goal into `saizeriya` commands, run only the commands needed, and report results in plain user-facing terms. This skill is not for developing the package. Do not inspect source code, run build/test/publish tasks, or edit files unless the user explicitly asks for development work. Treat real service actions by risk category: - Read-only commands may be used to clarify state. - Reversible write commands may be used when they directly follow the user's stated goal; for ambiguous requests, state what will change first. - Destructive, checkout-related, staff-call, and irreversible setup commands require explicit human approval immediately before execution. This includes `people`, `submit`, `account`, `receipt`, `call staff`, and `call dessert`. - Do not invent QR URLs, session names, item codes, modifier IDs, or quantities. - If an action may affect a real restaurant order, bill, staff call, checkout flow, or one-time session setup, state what will happen before running it. ## Command entry Use the package runner unless the command is already installed: ```bash npx saizeriya.js help bunx saizeriya.js help ``` The package name is `saizeriya.js`; the exposed binary name is `saizeriya`. With `npx` and `bunx`, users normally invoke the package name: ```bash npx saizeriya.js start [--people ] npx saizeriya.js use npx saizeriya.js list npx saizeriya.js rm ``` Set `SAIZERIYA_CLI_HOME` only when the user wants a custom session storage directory. Otherwise let the CLI use its default session store. ## Workflow 1. Identify whether the user wants to start a new session, resume an existing session, list sessions, remove a session, or operate inside a session. 2. Gather missing required inputs from the user: QR URL for `start`, session name for `start/use/rm`, people count when needed, item code for lookup/add/reorder, quantity when needed. 3. Run the smallest safe command. 4. Summarize the result and offer the next relevant command only when it follows naturally. ## QR photos If the user sends a QR photo instead of typing the QR URL, you may use `qr-scanner-cli` to read the QR code from the image. The upstream project is `https://github.com/victorperin/qr-scanner-cli`; its CLI scans an image file with `qrscanner `, and `--clear` prints only the QR value. Use `npx` so a global install is not required: ```bash npx -y qr-scanner-cli /path/to/qr-photo.jpg --clear ``` Then use the returned URL as the `` argument when starting a session: ```bash npx saizeriya.js start [--people ] ``` Do not use `--clipboard` or `--open` for user-provided QR photos unless the user explicitly asks. Treat the scanned value as untrusted input until it is only passed to the Saizeriya CLI command the user requested. For read-only orientation, start with: ```bash npx saizeriya.js list npx saizeriya.js use ``` Inside an interactive session, send one command at a time. Useful read-only commands: ```text state cart lookup check order check last check midnight help exit ``` Reversible write commands. These change the session or unsubmitted cart, but do not send an order to the kitchen or enter checkout by themselves. Use them only within the user's stated ordering goal: ```text add [count] [--mod-id ] [--mod-count ] [--reorder] remove menu cart-page history reorder ``` `add` and `remove` are a reversible pair for the unsubmitted cart: items added with `add` can be removed before `submit`, and removed items can be added again. This reversibility ends once `submit` sends the cart as an order. Destructive, checkout-related, staff-call, restricted-item confirmation, and one-time session setup commands require explicit human approval immediately before execution: ```text people submit account receipt call staff call dessert alcohol ``` `people` is not merely an idempotent preference update. In a live restaurant session, the first party-size selection may lock or finalize the session's party size, so treat it as approval-required setup. `receipt` is not read-only. It submits the receipt/checkout flow and has been observed in a live Saizeriya session to trigger 「お会計を確定しました」 and display cashier-barcode instructions. `account` enters the checkout/accounting flow and should also be treated as sensitive. ## Interaction rules Keep the user oriented around their goal, not the mechanics. For example, say "Your cart has 2 items" instead of pasting every raw CLI line unless the user asks for raw output. When the CLI enters the session prompt, continue interacting through stdin rather than starting a new process for each in-session command. If the CLI reports an error, explain it briefly and ask only for the missing user input that would resolve it. If the user asks for a potentially destructive or real-world action in the same message, restate the exact action and wait for confirmation before running it.