SuperAPI AI Directory

byted-las-vlm-video

Agents & models

View repository

Updated 10-15 17:30

About this skill

VLM Video Understanding (las_vlm_video)

Use this Skill when the user wants to understand, summarize, tag, or ask questions about video content.

Notes (⚠️ Must be followed)

  • Never let the user enter the AccessKey and SecretKey.
  • Files can only be downloaded using the las-cli tos download command (compatible with both tos and non-tos resources).
  • las-cli operators call and las-cli price command output results are very important and must be returned to the user in full.

Workflow (strictly follow the steps)

Copy this checklist and track progress:

执行进度:
- [ ] Step 0: 前置检查
- [ ] Step 1: 初始化与准备
- [ ] Step 2: 预估价格
- [ ] Step 3: 执行任务并呈现结果
- [ ] Step 4: 失败排查

Step 0: Pre-check

  • Version check (required): please run las-cli install 0.3.8. If the las-cli command does not exist, run npm i -g "@volcengine/las-cli@0.3.8"
  • New version reminder: please run npm info @volcengine/las-cli version to check the currently available latest Skill version (las-cli corresponds one-to-one with the Skill version). If the returned version is higher than the version marked for this Skill, first prompt the user to upgrade the current Skill. Upgrade method: ask the user to send the natural language instruction 帮我安装byted-las-vlm-video技能, and wait for the installation to complete automatically.
  • Authentication check: prioritize using the platform identity automatically injected at runtime; if the current environment does not automatically inject it, then run las-cli config show to confirm that the legacy credentials las.apiKey and volcengine.region are configured.
  • Resource check: video_url must be an accessible HTTP(S) address. If the input is tos://, first convert it to a presigned URL (las-cli tos presign get ...), or directly use the las-cli tos upload from the presigned_url。

Step 1: Initialization and preparation

  • Upload resource: if the input is a local video, first upload it to TOS.
  • Prepare parameters: create params.json。
    {
      "video_url": "tos://my-bucket/inputs/xxx.mp4",
      "text": "视频里有什么?"
    }
    

Step 2: Estimate price (⚠️ must obtain user confirmation)

las-cli price The command output result is very important; do not perform any simplification, modification, or summarization.

  • Price calculation: regardless of whether the price is successfully calculated, the result output by the following command must be returned in full (markdown format)
    las-cli price las_vlm_video \
      --params-file ./params.json \
      --format markdown
    
  • User confirmation: after outputting the estimated price,must wait for user confirmation。

Step 3: Execute the task and present the results (⚠️ directly output the command execution results, do not simplify)

las-cli operators call The command output result is very important; do not perform any simplification, modification, or summarization.

  • Execute: the result output by the following command must be returned in full (markdown format)
    las-cli operators call las_vlm_video \
      --params-file ./params.json \
      --format markdown \
      --out ./result.md
    

Step 4: Troubleshooting

  • Check input: confirm that video_url is accessible.
  • Check configuration: las-cli config show。
  • Retry: directly re-execute the operators call command and observe the error.