AI 生产力工具助手

v1.0.0

批量生成视频图片客户端助手,可以通过对话驱动本地客户端生成视频和图片。 支持对接飞书、QQ、微信等对话入口,不用再限制在电脑前使用,随时随地都可以生成。 当用户提到以下意图时使用此技能: 「生成视频」「生成图片」「文生视频」「图生视频」「文生图」「图生图」 「Sora 生成」「Veo 生成」「Nano Banan...

0· 141·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for xiaolou888/ai-productivity-assistant.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "AI 生产力工具助手" (xiaolou888/ai-productivity-assistant) from ClawHub.
Skill page: https://clawhub.ai/xiaolou888/ai-productivity-assistant
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install ai-productivity-assistant

ClawHub CLI

Package manager switcher

npx clawhub@latest install ai-productivity-assistant
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description claim a dialog-driven local client wrapper for bulk video/image generation; the included SKILL.md and scripts/local_client.py only reference a local HTTP service at 127.0.0.1:17321 and the requests library. Declared dependency (requests) matches usage. No unrelated credentials, binaries, or external URLs are required by the skill.
Instruction Scope
Runtime instructions restrict all calls to http://127.0.0.1:17321 and instruct the agent to return task IDs (avoid continuous polling). Note: the skill intentionally delegates external network forwarding to the local DYU client (the client may itself be configured with a gateway URL), so while the skill does not contact external services directly, submitted tasks or references (image_url, metadata.urls) can cause the local client to fetch or forward to external AI services. This delegation is expected given the described architecture but is worth understanding before use.
Install Mechanism
Instruction-only skill with an included Python module. There is no installer or remote download. The only declared dependency is requests, which is proportionate to the code that performs local HTTP calls.
Credentials
The skill declares no required environment variables, no primary credential, and no config paths. The code does not attempt to read environment secrets. This is proportionate to a local-client wrapper that requires only a running local service.
Persistence & Privilege
Skill is not always-enabled, does not request persistent elevated privileges, and contains no code that modifies other skills or system-wide configurations. Autonomous invocation is allowed by platform default but not combined with other concerning privileges.
Assessment
This skill appears internally consistent: it only talks to a local service at 127.0.0.1:17321 and does not ask for API keys or other credentials. Before installing, confirm the DYU/local client running on that port is from a trusted source and review its gateway configuration because the local client — not the skill — may forward tasks to external AI services. Avoid pasting secrets or private/internal-only URLs into prompts or metadata.urls, and consider running the local client in an isolated environment if you are concerned about data leaving your machine.

Like a lobster shell, security has layers — review code before you run it.

latestvk978vcz10cxw6h719s1xyep91h836c9p
141downloads
0stars
1versions
Updated 1mo ago
v1.0.0
MIT-0

AI 生产力工具助手

一、必读约束

工作原理

本技能与本地已安装的视频/图片批量生成客户端通信,无需 API Key,无需配置凭证。

你(对话) → OpenClaw Skill → http://127.0.0.1:17321 → 本地批量生成客户端 → AI 生成服务

前置条件

本地批量生成客户端必须处于运行状态,且已完成 SK 配置。每次执行任务前,先调用 scripts/local_client.py 中的 check_client() 确认客户端在线。

任务处理规范

  1. 提交成功后必须立即返回 task_id,不要在对话里持续轮询刷屏(不要调用 wait_for_result
  2. 提交后统一提示用户:生成结果大概需要 3-10 分钟,请稍后查询结果
  3. 给出“一键复制”的查询指令:必须单独放在一个 text 代码块中(便于用户点击复制)
    • 示例:
      查询任务 task_xxxxxxxxxxxxx
      
  4. 只有当用户明确要求「查询进度 / 查询结果 / 等结果」时,才调用查询接口:
    • 默认只查询 一次GET /api/task/{task_id}
    • 如用户要求持续跟进,可按用户要求轮询,但要控制频率(建议 5 秒一次)
    • 拿到结果链接时,必须使用 video_url(首选)或 url,并且原样保留 ? 之后的全部查询参数
  5. status = completed → 将 url 字段展示给用户(图片直接渲染,视频给出链接)
  6. status = failed → 告知用户失败原因,并询问是否重新生成

安全约定

  • 所有请求只发往 http://127.0.0.1:17321,不访问任何外部地址
  • 客户端离线时,友好提示用户启动 DYU 客户端

二、本地接口说明

所有接口均为 http://127.0.0.1:17321,调用 scripts/local_client.py 封装的客户端。

from local_client import get_client

client = get_client()

# 检查客户端状态
status = client.status()

# 提交单个任务
result = client.generate({ "model": "...", "prompt": "..." })

# 提交批量任务
results = client.generate_batch([
    { "model": "...", "prompt": "..." },
    { "model": "...", "prompt": "..." },
])

# 查询单个任务
task = client.query_task("task_xxxxx")

# 批量查询任务
tasks = client.query_tasks(["task_xxx1", "task_xxx2"])

# (可选)等待任务完成(自动轮询)
final = client.wait_for_result("task_xxxxx", timeout=600)

三、快速决策表

用户意图示例说法模型参数备注
视频 - Sora
Sora 横屏 10 秒"用 Sora 生成横屏视频"sora-2-landscape-10s默认横屏选此
Sora 竖屏 10 秒"生成竖屏短视频"sora-2-portrait-10s
Sora 横屏 15 秒"15 秒横屏视频"sora-2-landscape-15s
Sora 竖屏 15 秒"15 秒竖屏视频"sora-2-portrait-15s
Sora Pro 横屏 25 秒"25 秒高质量视频"sora-2-pro-landscape-25s
Sora Pro 竖屏 25 秒"25 秒竖屏高质量"sora-2-pro-portrait-25s
Sora Pro HD 横屏 15 秒"HD 高清横屏视频"sora-2-pro-landscape-hd-15s
Sora Pro HD 竖屏 15 秒"HD 高清竖屏视频"sora-2-pro-portrait-hd-15s
Sora 图生视频(URL)"把这张图做成视频 https://..."sora-2-portrait-10s + image_url
视频 - Veo
Veo 文生视频"用 Veo 生成视频"veo_3_1-fast + size=1920x1080标清/默认
Veo 竖屏"Veo 竖屏视频"veo_3_1-fast + size=1080x1920标清/默认
Veo 高清文生视频"用 Veo 生成高清视频"、"Veo 高清 1080P"veo_3_1-fast-hd + size=1920x1080需在 /api/pricing 中已启用
Veo 高清竖屏"Veo 高清竖屏视频"veo_3_1-fast-hd + size=1080x1920需在 /api/pricing 中已启用
Veo 首尾帧"以这两张图为首尾帧"veo_3_1-fast-fl + size=1920x1080需要 image_url
Veo 高清首尾帧"以这两张图为首尾帧生成高清视频"veo_3_1-fast-fl-hd + size=1920x1080需要 image_url,需在 /api/pricing 中已启用
Veo 参考图"参考这张图生成视频"veo_3_1-fast + input_reference
图片 - Nano Banana
标准文生图"生成一张图片"nano_banana_2默认选此
Pro 文生图"生成 Pro 图片"nano_banana_pro
1K 高清图"生成 1K 高清图"nano_banana_pro-1K
2K 超清图"生成 2K 超清图"nano_banana_pro-2K
4K 极清图"生成 4K 极清图"nano_banana_pro-4K
图生图"把这张图转成油画"nano_banana_2 + metadata.urls

四、意图关键词判断

横竖屏判断

用户说参数
横屏、横版、16:9、landscapeSora 选 landscape 模型;Veo 选 size=1920x1080
竖屏、竖版、9:16、portraitSora 选 portrait 模型;Veo 选 size=1080x1920
未指定默认横屏

时长判断(Sora)

用户说模型后缀
10 秒、10s、短视频-10s
15 秒、15s-15s
25 秒、25s、长视频-25s(Pro)
HD、高清-hd-15s(Pro HD)
未指定默认 -10s

图片分辨率判断(Nano Banana)

用户说模型
普通、标准、快速nano_banana_2
Pronano_banana_pro
1Knano_banana_pro-1K
2Knano_banana_pro-2K
4Knano_banana_pro-4K
未指定nano_banana_2

宽高比判断(图片)

用户说metadata.aspectRatio
横图、横版、16:916:9
竖图、竖版、9:169:16
正方形、1:11:1
未指定auto

五、单个生成示例

示例 1:Sora 文生视频

from local_client import get_client
client = get_client()

result = client.generate({
    "model": "sora-2-landscape-10s",
    "prompt": "一只可爱的橘猫在阳光明媚的花园里玩耍,镜头缓慢推进"
})
# result = {"success": True, "task_id": "task_xxx", "status": "queued", ...}
# 对话中不要轮询:直接把 task_id 返回给用户,并提示 3-10 分钟后查询:
# 你可以发送:查询任务 task_xxx

示例 2:Veo 文生视频

result = client.generate({
    "model": "veo_3_1-fast",
    "prompt": "夕阳下的海边,海浪轻轻拍打礁石,光影变幻",
    "size": "1920x1080"
})
# 对话中不要轮询:直接返回 task_id,并提示稍后查询

示例 3:Sora 图生视频(图片 URL)

result = client.generate({
    "model": "sora-2-portrait-10s",
    "prompt": "让画面动起来,人物微微转头微笑",
    "image_url": "https://example.com/photo.jpg"
})
# 对话中不要轮询:直接返回 task_id,并提示稍后查询

示例 4:Nano Banana 文生图

result = client.generate({
    "model": "nano_banana_2",
    "prompt": "美丽的日出风景,金色阳光洒在宁静湖面,远处连绵山脉,写实风格",
    "metadata": {
        "aspectRatio": "16:9",
        "urls": []
    }
})
# 对话中不要轮询:直接返回 task_id,并提示 3-10 分钟后查询:
# 你可以发送:查询任务 task_xxx

示例 5:Nano Banana 图生图

result = client.generate({
    "model": "nano_banana_2",
    "prompt": "将这张图片转换成梵高油画风格,保留主体构图",
    "metadata": {
        "aspectRatio": "1:1",
        "urls": ["https://example.com/reference.jpg"]
    }
})
# 对话中不要轮询:直接返回 task_id,并提示稍后查询

六、批量生成示例

示例 6:批量文生视频(Sora,多个提示词)

from local_client import get_client
client = get_client()

tasks = [
    {
        "model": "sora-2-landscape-10s",
        "prompt": "春天的樱花树下,花瓣随风飘落"
    },
    {
        "model": "sora-2-landscape-10s",
        "prompt": "夏日海边,孩子们在沙滩上奔跑嬉戏"
    },
    {
        "model": "sora-2-landscape-10s",
        "prompt": "秋天的枫叶林,金红色叶片在阳光下闪耀"
    },
    {
        "model": "sora-2-landscape-10s",
        "prompt": "冬日雪景,白雪覆盖的山村宁静祥和"
    }
]

# 提交批量任务
batch_result = client.generate_batch(tasks)
# batch_result = {"success": True, "results": [{"task_id": "task_001", ...}, ...]}

# 提取所有 task_id
task_ids = [r["task_id"] for r in batch_result["results"] if r.get("success")]

# 对话中不要轮询:直接返回 task_ids,并提示用户稍后逐个查询:
# 你可以发送:查询任务 task_001

示例 7:批量文生图(Nano Banana,不同比例)

tasks = [
    {
        "model": "nano_banana_pro-1K",
        "prompt": "产品主图:一款高端智能手表,白色背景,专业摄影",
        "metadata": {"aspectRatio": "1:1", "urls": []}
    },
    {
        "model": "nano_banana_pro-1K",
        "prompt": "产品场景图:智能手表佩戴在手腕上,户外运动场景",
        "metadata": {"aspectRatio": "16:9", "urls": []}
    },
    {
        "model": "nano_banana_pro-1K",
        "prompt": "产品细节图:智能手表表盘特写,展示精致工艺",
        "metadata": {"aspectRatio": "9:16", "urls": []}
    }
]

batch_result = client.generate_batch(tasks)
task_ids = [r["task_id"] for r in batch_result["results"] if r.get("success")]
# 对话中不要轮询:直接返回 task_ids,并提示稍后查询

示例 8:批量图生图(风格迁移)

reference_images = [
    "https://example.com/photo1.jpg",
    "https://example.com/photo2.jpg",
    "https://example.com/photo3.jpg",
]

tasks = [
    {
        "model": "nano_banana_2",
        "prompt": "转换成水彩画风格,保留原图构图和色调",
        "metadata": {"aspectRatio": "auto", "urls": [img_url]}
    }
    for img_url in reference_images
]

batch_result = client.generate_batch(tasks)
task_ids = [r["task_id"] for r in batch_result["results"] if r.get("success")]
# 对话中不要轮询:直接返回 task_ids,并提示稍后查询

示例 9:批量 Veo 视频(不同尺寸)

tasks = [
    {
        "model": "veo_3_1-fast",
        "prompt": "城市夜景延时摄影,霓虹灯光流动",
        "size": "1920x1080"   # 横屏
    },
    {
        "model": "veo_3_1-fast",
        "prompt": "城市夜景延时摄影,霓虹灯光流动",
        "size": "1080x1920"   # 竖屏
    }
]

batch_result = client.generate_batch(tasks)
task_ids = [r["task_id"] for r in batch_result["results"] if r.get("success")]
# 对话中不要轮询:直接返回 task_ids,并提示稍后查询

七、查询任务示例

查询单个任务

task = client.query_task("task_xxxxxxxxxxxxx")
# {"success": True, "id": "task_xxx", "status": "completed", "url": "https://..."}

批量查询任务

tasks = client.query_tasks(["task_001", "task_002", "task_003"])
for t in tasks["results"]:
    print(f"{t['task_id']}: {t['status']} - {t.get('url', t.get('error', ''))}")

查询本地任务列表

# 查询所有任务(最近 50 条)
all_tasks = client.list_tasks()

# 按状态过滤
completed = client.list_tasks(status="completed")
processing = client.list_tasks(status="processing")

任务统计示例(按天/按类型/成功数量)

from datetime import date
from local_client import get_client

client = get_client()

# 统计今天的任务(按 Veo / Sora / 图片 分类)
today = date.today()
summary = client.summarize_tasks(target_date=today, status=None)

# summary 结构示例:
# {
#   "total": 12,
#   "by_type": {
#     "veo":   {"total": 5, "success": 4, "failed": 1, "ids": ["task_xxx", ...]},
#     "sora":  {"total": 3, "success": 3, "failed": 0, "ids": [...]},
#     "image": {"total": 4, "success": 2, "failed": 2, "ids": [...]}
#   }
# }

对话说明:当用户说「我想查询一下今天生成了多少条任务,veo 有多少,sora 有多少,图片有多少,成功有多少,把成功的 ID 发我一下,分类发我」时:

  1. 优先调用 client.summarize_tasks(target_date=今天, status=None) 拿到结构化统计。
  2. 只统计本地任务表中 video_idtask_ 开头的记录(严格视为 upstream 任务 ID)。
  3. 按 Veo / Sora / 图片 三类分别用自然语言说明总数、成功/失败数。
  4. 把成功的 ID 按类型分别列出,例如:
    • Veo 成功任务 ID:
    task_xxxxxxxx
    task_yyyyyyyy
    
    • Sora 成功任务 ID:
    task_aaaaaaaa
    
    • 图片成功任务 ID:
    task_bbbbbbbb
    

八、结果展示规范

  • 图片/视频链接(关键):链接必须100%复用接口返回的原始字符串(尤其包含 ?Expires=...&Signature=... 的临时签名链接),不得做任何裁剪/解码/重写/格式化
  • 绝对禁止输出 Markdown 链接:[点击查看](URL)(QClaw 可能会把 ? 后参数截断,导致链接失效)。
  • 唯一允许的输出方式:把接口返回的 video_url(首选)或 url 原样放进 text 代码块,便于“一键复制”,并确保不会丢参数:
https://...完整URL(包含 ? 与 & 之后的全部参数,原样粘贴)
  • (可选)如果你要提供可点击链接,只能额外给一行尖括号自动链接,仍必须是原样完整 URL:<https://...>
  • 批量结果:以表格或列表形式展示每个任务的状态和结果
  • 失败任务:明确告知失败原因,询问用户是否重新生成
  • 查询指令:必须用单独的 text 代码块输出,便于一键复制,例如:
查询任务 task_xxxxxxxxxxxxx

九、功能模块详情

Comments

Loading comments...