group-director#

Use this skill only as a video execution layer for Feishu/Lark group-chat video generation.

Hard rules#

• Assume Claw already has enough context from its own memory and recent reading.
• Do not ask extra questions in normal group-director use.
• Do not ask the user to repeat or recap the group chat.
• Do not read Feishu/Lark history inside this skill.
• Do not summarize group chat inside this skill.
• Treat final_video_prompt as already finalized by Claw.
• Use exactly two steps:
  1. create the task
  2. poll by Python every 30 seconds until it finishes
• Keep the model fixed to Seedance-Pro-1.5.
• Always set provider_specific.generate_audio=true.
• Never send raw JSON back to Feishu.
• When the task finishes successfully, send only a normal natural-language message plus the final video_url.
• Do not send debug objects, raw provider payloads, or JSON-looking blocks to Feishu.

What Claw is responsible for#

Claw is responsible for:

• reading the group context
• using its own memory and recent context
• summarizing the discussion when needed
• turning that into the final video prompt
• calling this skill
• sending the final natural-language message back to Feishu/Lark

What this skill is responsible for#

This skill is responsible only for:

• creating the video task
• polling status until completion, failure, or timeout
• returning the final video URL or a plain-text failure message

Fixed defaults#

• model: Seedance-Pro-1.5
• duration: 12
• resolution: 720p
• allowed orientation: portrait or landscape
• poll interval: 30 seconds
• timeout: 600 seconds
• audio generation: always enabled

Do not expose square mode. Do not let the model vary. Do not omit generate_audio.

Input expected from Claw#

Required:

• final_video_prompt

Optional:

• orientation

If orientation is missing, default to portrait. If orientation is present, it must be portrait or landscape.

Calling pattern#

Step 1: create task#

bash

python3 scripts/main.py video-create \
  --final-video-prompt "这里放 Claw 已整理好的最终视频提示词" \
  --orientation portrait

This returns only the task_id as plain text.

Step 2: poll by Python until done#

bash

python3 scripts/main.py video-poll --task-id "task_xxx"

This polls every 30 seconds until one of these happens:

• completed -> prints only the final video_url
• failed -> prints a plain-text failure message
• timeout -> prints a plain-text timeout message

Feishu return rule#

When Claw sends the result back to Feishu/Lark:

• use a normal message
• include the final video_url
• do not include raw JSON
• do not include provider payloads
• do not paste structured blobs

Good style:

text

视频生成好了：
https://...

Bad style:

text

{"status":"completed","task_id":"...","video_url":"https://..."}

Environment variable#

Only this variable is required:

bash

export SENSEAUDIO_API_KEY="your_key"

Optional:

bash

export SENSEAUDIO_BASE_URL="https://api.senseaudio.cn"

Resource layout#

• scripts/main.py: CLI entrypoint with plain-text outputs
• scripts/video_api.py: provider wrapper for create, status, and polling
• references/provider_notes.md: fixed model and parameter notes
• references/integration_cn.md: Chinese integration rules for Claw