Xiaoyingma
Start creatingStart

Home / Features / Agent CLI

Agent CLI

Bring Xiaoyingma into your Agent

One prompt brings Xiaoyingma into your Agent workflow.

Choose an installation method

Send this prompt to your AI Agent

It downloads the Skill package, completes local setup, and follows the guide to run the environment check first.

Prompt

Please install the Xiaoyingma Agent CLI Skill: https://yingma.tv/downloads/yingma-cli-skill-0.2.0.zip

Restart your AI Agent after setup, then you are ready to begin.

Supported AI Agents

Use in sessions that can download files and install local Skills.

  • Codex
  • Claude Code
  • OpenClaw
  • Hermes
  • WorkBuddy
  • 豆包

How the Skill works

The Skill has the Agent inspect first, then provide the smallest plan and a preview; it stops for your approval before writes, generation, or refinements.

Use $yingma-cli: inspect the current project and canvas first, then give me the smallest change plan and a --dry-run. Do not modify, submit a model task, or render until I confirm.

Upload the command pack in My skills

Sign in when you approve

Browser sessions do not sync to the CLI. Start verification-code sign-in only after you explicitly ask; the token is stored only in your local configuration.

  1. 01

    Get a captcha image

    yingma auth captcha --output ./yingma-captcha.png
  2. 02

    Send an email code

    yingma auth request-email --email YOU@example.com --captcha-id ID --captcha-code CODE
  3. 03

    Save the local session

    yingma auth login-email --email YOU@example.com --code 123456

Operate projects safely

Read first. Changes to projects or credit-consuming actions require clearer confirmation.

Check the environment first

yingma doctor

Check only the local CLI, sign-in state, and directory binding. It does not read browser credentials or request project data.

Validate the canvas first

yingma canvas validate

Find duplicate IDs, dangling links, and invalid nodes before import.

Preview writes first

--dry-run

Preview the request target and parameters, then create, generate, or render only after confirmation.

AGENT SCOPE

What an Agent can do through the CLI

State the creative goal, assets, and approval points clearly; the Agent organizes the work into inspectable projects and canvases.

You want toStart from one idea

The Agent can use the CLI toCreate a project, organize a script draft and shot structure, then build an inspectable canvas plan.

You want toRecreate a video style

The Agent can use the CLI toUpload your reference video or images and create reference and generation nodes without reusing branded source elements.

You want toMake a short drama or animation

The Agent can use the CLI toOrganize an episode canvas with script, characters, scenes, keyframes, and video nodes, including dependencies.

You want toMake a product commercial

The Agent can use the CLI toUpload product images and plan the commercial script, visual references, shots, and video generation nodes.

You want toTrack production

The Agent can use the CLI toRead project, canvas node, asset, and episode lists to identify nodes that need follow-up.

You want toContinue an existing project

The Agent can use the CLI toRead the existing canvas, propose precise node and edge edits, and preview the change first.

You want toRefine an existing video

The Agent can use the CLI toInspect the target video node, then preview prompt optimization, reshoot, trim, split, HD, erase, or depth work before submitting a task.

You want toPrepare delivery

The Agent can use the CLI toReview assets and episode lists, then export canvas state for backup, comparison, or handoff.

SEND TO YOUR AGENT

Use it like this

After installation, choose sign-in yourself. Once signed in, send a task below to your Agent. It will report a plan and preview before waiting for execution approval.

01
Product commercial plan

Task for your Agent

I will provide a reference commercial and a product image.

Use $yingma-cli to create a product-commercial project for a 30-second vertical video.

Requirements:
1. Analyze the reference pacing, shot structure, transitions, product timing, and emotional curve;
2. Replace the product with my supplied image and do not reuse branded source elements;
3. Build canvas nodes for the ad script, shot breakdown, product visual references, keyframes, and video generation;
4. State each video node’s reference usage, duration, camera movement, and product exposure;
5. Create only the project and complete canvas first. Return the project ID, canvas ID, node summary, and a generation preview. Do not start image or video generation, or rendering, until I confirm.
02
Short-drama episode canvas

Task for your Agent

Use $yingma-cli to create an inspectable project and episode canvas for the first episode of a short drama. The theme is:
“An intern ignored by the whole company suddenly discovers that she can see the first lie each person will tell today.”

Requirements:
1. Draft a 60–90 second story with an opening hook, protagonist goal, resistance, reversal, and closing hook;
2. Split it into 6–10 filmable video units, each with characters, scene, action, and dialogue purpose;
3. Organize script, characters, scenes, keyframes, and video nodes in the canvas and connect their dependencies;
4. Return the episode structure, node summary, and a generation plan. Do not start image or video generation, or rendering, until I confirm.
03
Continue the current project

Task for your Agent

Use $yingma-cli to inspect the current project.

First read the project, current canvas, nodes, assets, and episode status. Find failed nodes, dangling edges, incomplete prompts, or videos that can be refined. Then:
1. Explain each issue and the smallest safe repair;
2. List the nodes or edges that would need a change;
3. For prompt optimization or video reshoot, trim, and HD work, prepare --dry-run previews only;
4. Do not modify, rerun nodes, generate, or render;
5. Return a project-status summary and the next step that needs my confirmation.

Creating a project or canvas writes to your account. Each example tells the Agent to wait before generation, rendering, or an actual edit. Replace the theme, assets, and duration as needed.