← Back to Skills Marketplace
fandywang87

AI RecSys Weekly Report

by fandywang87 · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
116
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install ai-recsys-weekly-report
Description
自动生成关于"AI大模型(LLM、VLM等)技术在搜索、广告、推荐(搜广推)领域应用"的深度技术周报, 并自动同步到 IMA(腾讯文档/知识库)。触发词:AI搜广推技术周报、搜广推周报、大模型推荐报告、 推荐系统技术周报、生成式推荐周报、Transformer推荐系统报告、Scaling Law推荐系统、 MoE...
README (SKILL.md)

AI 搜广推技术周报 - 自动化 Skill

本 Skill 将完整的"信息搜集 → 报告生成 → IMA 同步"流程打包,可迁移到任意 WorkBuddy 账号使用。

前置依赖

必须安装的 Skill

本 Skill 依赖以下 Skill,使用前请确保已安装:

  1. ima-skills(或 腾讯ima)— 用于上传文件到 IMA 知识库
    • 安装后需配置 Client ID 和 API Key(见下方「IMA 配置」)

可选但推荐的 Skill

  • wechat-article-search — 搜索微信公众号文章
  • zhihu-search-api-skill — 搜索知乎文章
  • ArXiv论文追踪 — 搜索 ArXiv 最新论文

IMA 配置

在使用此 Skill 前,必须先配置 IMA 凭证:

# 1. 打开 https://ima.qq.com/agent-interface 获取 Client ID 和 API Key

# 2. 存储凭证
mkdir -p ~/.config/ima
echo "你的Client_ID" > ~/.config/ima/client_id
echo "你的API_Key" > ~/.config/ima/api_key

使用方式

方式一:单次执行(手动触发)

当用户要求"生成搜广推周报"、"出一份大模型推荐系统的技术报告"时,按以下步骤执行:

Step 1:信息搜集

并行执行以下搜索(至少覆盖 ArXiv + 中文来源):

ArXiv 论文搜索(核心):

  • site:arxiv.org recommendation system transformer backbone 2025 2026
  • site:arxiv.org recommendation MoE mixture of experts ranking
  • site:arxiv.org sparse attention recommendation transformer
  • site:arxiv.org generative recommendation LLM

中文技术文章搜索

  • 微信公众号:大模型推荐系统 Transformer 搜广推 (通过 wechat-article-search 或 web_search)
  • 知乎:搜广推 Scaling Law 推荐系统 大模型 (通过 zhihu-search 或 web_search)
  • 技术博客:InfoQ、腾讯云开发者社区、CSDN 等

特定技术追踪

  • DeepSeek Sparse Attention (DSA)
  • Kimi Attention Residuals (AttnRes)
  • Qwen Gated Attention
  • Muon 优化器
  • SwiGLU 激活函数
  • DeepSeekMoE

对搜索到的重要论文,使用 web_fetch 获取 arXiv 页面的详细信息(标题、作者、摘要、核心贡献)。

Step 2:生成报告

将搜集到的信息整理为 Markdown 格式的技术周报,必须包含以下章节:

  1. 本周热点综述(3-5条核心趋势,每条含论文链接)
  2. 深度技术解读(选择2-5篇最重要的工作,每篇包含):
    • 论文基本信息(标题、作者、机构、会议/期刊、链接)
    • 核心创新点
    • 关键实验数据(具体数字)
    • 对搜广推领域的启示和建议
  3. 技术对比分析
    • 至少一个对比表格(如不同架构对比、不同注意力机制对比)
    • 可选 ASCII 架构图展示演进路线
  4. 创新思考与建议(2-4条可操作的建议)
  5. 趋势预测(2-4个方向的前瞻性判断)
  6. 延伸阅读推荐(6-12条,分类整理:综述类 / 经典必读 / 技术博客)
  7. 优化调整思考建议(新增!本次研究的局限与后续改进方向)
  8. 参考文献(所有引用的完整列表,每项含可点击链接)

格式要求

  • 每篇论文/工作必须附带 arXiv 链接和 PDF 直接下载链接
  • 尽可能附上中文解读链接(知乎/CSDN/博客园等)
  • 如有 GitHub 开源仓库,附上 GitHub 链接
  • 报告字数:2000-4000 字
  • 文件命名:AI搜广推技术周报_YYYY-MM-DD.md

Step 3:上传到 IMA 知识库

关键:必须使用 cos-upload.cjs 官方脚本上传 COS,不可用 curl 手动传。

标准三步流程:

# 配置常量(每次执行时替换实际值)
FILE_PATH="\x3C生成的报告路径>"
FILE_NAME="AI搜广推技术周报_YYYY-MM-DD.md"
FILE_EXT="md"
KB_ID="\x3C目标知识库ID>"          # 用户需提供
MEDIA_TYPE=7                     # Markdown 固定为7
CONTENT_TYPE="text/markdown"

# 工具路径
NODE="/Users/fandywang/.workbuddy/binaries/node/versions/22.12.0/bin/node"
# 注意:目标账号需要确认 node 路径,可能不同
COS_SCRIPT="\x3Cskill目录>/scripts/upload-to-ima.py"

# 执行上传脚本
python3 "$COS_SCRIPT" --file "$FILE_PATH" \
  --kb-id "$KB_ID" \
  --title "$FILE_NAME"

或直接调用 scripts/upload-to-ima.py 封装脚本(见下方脚本说明)。

方式二:自动化定时任务

创建每周定时执行的自动化任务:

名称: AI大模型搜广推技术周报
频率: FREQ=WEEKLY;BYDAY=MO;BYHOUR=9;BYMINUTE=0
prompt: 请加载 ai-recsys-weekly-report skill,完整执行一次周报生成并同步到 IMA 知识库。
      知识库ID: \x3C用户提供>
cwds: \x3C用户的工作空间路径>
status: ACTIVE

脚本说明

scripts/upload-to-ima.py

IMA 知识库上传封装脚本,内部实现标准的三步流程:

  1. create_media — 获取 media_id 和 COS 上传凭证
  2. cos-upload.cjs — 使用腾讯云 SDK 上传文件到 COS
  3. add_knowledge — 将已上传文件关联到知识库

用法

python3 scripts/upload-to-ima.py --file \x3C报告路径> --kb-id \x3C知识库ID> [--title \x3C自定义标题>]

依赖

  • Python 3.x
  • ima-skills 中的 cos-upload.cjs 脚本
  • Node.js(用于运行 cos-upload.cjs)

重要注意事项

  • 如果 cos-upload.cjs 返回非零退出码,立即终止流程,不要继续调用 add_knowledge
  • Node 路径需要在脚本中配置正确(不同环境可能路径不同)
  • IMA 凭证从 ~/.config/ima/client_id~/.config/ima/api_key 读取

研究范围参考

以下是本 Skill 覆盖的核心技术方向,供搜索时参考关键词:

技术方向 代表性工作 搜索关键词
统一 Backbone OneTrans, RankMixer, MixFormer unified backbone recommender, one transformer
Scaling Law 10亿参数推荐Transformer scaling law recommendation system
MoE 架构 DeepSeekMoE, 专家路由 mixture of experts recommendation
稀疏注意力 BlossomRec, DSA, ISA sparse attention sequential recommendation
残差连接革新 Kimi AttnRes attention residuals transformer
门控注意力 Qwen Gated Attention gated attention mechanism
优化器迁移 Muon, Adam-mini efficient optimizer recommendation
激活函数 SwiGLU, GeGLU activation function recommender
生成式推荐 GRU4Rec, GenRec, MTGR generative recommendation LLM
Usage Guidance
This skill appears internally consistent for generating reports and syncing them into Tencent IMA, but do these checks before installing: - Only install if you trust the ima-skills provider (cos-upload.cjs) because the Python wrapper executes that Node script with upload credentials. - Verify the cos-upload.cjs script content (from ima-skills) so it doesn't leak or mishandle credentials. - Store IMA client_id and api_key in a secure location and limit their privileges in IMA if possible; the skill expects them in ~/.config/ima. - Be aware the script prints subprocess stdout/stderr; run a test with a non-sensitive sandbox KB to confirm no secrets are logged. - If you plan automated weekly runs, ensure the WorkBuddy account and target KB have appropriate access controls and rotate API keys as part of your normal security hygiene.
Capability Analysis
Type: OpenClaw Skill Name: ai-recsys-weekly-report Version: 1.0.0 The skill bundle automates the generation of technical reports on AI recommendation systems and their synchronization with the Tencent IMA platform. The core logic in `scripts/upload-to-ima.py` manages a standard three-step upload process (media creation, COS upload, and knowledge base association) using the legitimate `ima.qq.com` API. While the skill reads credentials from a local config file (`~/.config/ima/`) and executes Node.js via `subprocess.run`, these actions are clearly aligned with the stated purpose of the tool. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found; the hardcoded environment-specific paths (e.g., for Node.js) appear to be usability flaws rather than intentional vulnerabilities.
Capability Tags
requires-sensitive-credentials
Capability Assessment
Purpose & Capability
The skill generates research reports and uploads them to IMA; it includes a helper upload script and explicitly depends on an IMA-related skill. Requested artifacts (IMA client_id/api_key stored in ~/.config/ima) and the included upload script are coherent with the described upload capability.
Instruction Scope
Runtime instructions are focused on web search (ArXiv, Chinese tech blogs) and report assembly, then uploading to IMA. The skill reads IMA credentials from ~/.config/ima and uses a bundled Python wrapper to call IMA APIs and an external cos-upload.cjs script; these behaviors match the upload task. Note: the process runs an external Node script (cos-upload.cjs) supplied by the dependent ima-skills; ensure that script is trusted.
Install Mechanism
There is no remote install/download step in the skill bundle (instruction-only plus a local Python script). No third-party URLs or archive extracts are included in the skill itself, so installation risk is low. The skill does require the separate ima-skills component to supply cos-upload.cjs.
Credentials
Only IMA credentials (client_id and api_key stored under ~/.config/ima) are required, which is proportionate to uploading to IMA. Minor concern: the upload wrapper logs subprocess stdout/stderr which might leak transient COS credentials if the underlying cos-upload.cjs prints them; review cos-upload.cjs and consider running in a controlled environment to avoid accidental credential exposure.
Persistence & Privilege
The skill does not request permanent or system-wide presence (always:false). It does not modify other skills or agent global settings in the provided files.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install ai-recsys-weekly-report
  3. After installation, invoke the skill by name or use /ai-recsys-weekly-report
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release: Auto-generate weekly tech reports on AI/LLM/VLM applications in search, advertising, and recommendation systems (搜广推). Includes ArXiv paper tracking, WeChat article search, IMA knowledge base sync, and standardized COS upload via cos-upload.cjs.
Metadata
Slug ai-recsys-weekly-report
Version 1.0.0
License MIT-0
All-time Installs 1
Active Installs 1
Total Versions 1
Frequently Asked Questions

What is AI RecSys Weekly Report?

自动生成关于"AI大模型(LLM、VLM等)技术在搜索、广告、推荐(搜广推)领域应用"的深度技术周报, 并自动同步到 IMA(腾讯文档/知识库)。触发词:AI搜广推技术周报、搜广推周报、大模型推荐报告、 推荐系统技术周报、生成式推荐周报、Transformer推荐系统报告、Scaling Law推荐系统、 MoE... It is an AI Agent Skill for Claude Code / OpenClaw, with 116 downloads so far.

How do I install AI RecSys Weekly Report?

Run "/install ai-recsys-weekly-report" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is AI RecSys Weekly Report free?

Yes, AI RecSys Weekly Report is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does AI RecSys Weekly Report support?

AI RecSys Weekly Report is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created AI RecSys Weekly Report?

It is built and maintained by fandywang87 (@fandywang87); the current version is v1.0.0.

💬 Comments