广告合规性审查工具

(组合技能) 对电商素材（图片/视频）进行两阶段的合规性审查：API快筛通用风险，再由AI模型结合专属知识库进行广告法深度分析。

Security Scan

The skill's requirements and instructions are internally consistent with an ad-compliance reviewer that composes other skills for API/video checks; no disproportionate permissions or installs are requested, though you should verify the child skills it calls and confirm a referenced prompt file exists.

Purpose & Capability

The name/description (two-stage ad compliance for images/videos) matches the runtime instructions: it performs a fast API-based scan then an AI deep analysis using the included knowledge_base and prompt_template. It declares no binaries, env vars, or installs, which is proportionate for an instruction-only, composition skill.

Instruction Scope

Most instructions stay within the stated scope (call api-check for images/text, call video-check for videos, then use local prompt_template.txt and knowledge_base/ for AI analysis). Two minor issues: (1) it references ecommerce_prompt.txt 'in the video-check skill directory' which is not present in this skill's manifest — you should confirm that file exists in the video-check skill or otherwise provide it; (2) the template assumes OCR/audio-to-text extraction has happened ('从图片OCR或视频语音中提取') but does not state which tool/skill performs OCR or speech-to-text, so the agent's available tools or child skills will determine how that step runs.

Install Mechanism

There is no install spec and no code files executed at runtime; this is instruction-only content (prompt templates and knowledge base). That minimizes filesystem and network install risk.

Credentials

This skill itself requests no environment variables or credentials. However, it calls child skills (api-check and video-check) and mentions using 'Baidu short-video audit' via the video-check flow — those child skills may require API keys/credentials and network access. Verify those downstream requirements before use.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request permanent presence or privileged modifications to other skills. Autonomous invocation is allowed by platform default but not an added privilege here.

Assessment

This skill is largely coherent and instruction-only, but before installing: (1) Confirm the child skills it calls (api-check and video-check) are trusted — review their manifests, required credentials, and network endpoints, since those skills will actually submit media to external APIs. (2) Verify the referenced file ecommerce_prompt.txt exists in the video-check skill (or update the composition to provide it). (3) Decide how OCR/speech-to-text will be performed (which tool/skill will extract TEXT_TO_REVIEW) and confirm any associated data-sharing/privacy implications. (4) If you will process sensitive media, check that downstream APIs (e.g., Baidu) meet your data protection requirements and that appropriate credentials are scoped minimally.

Ad Compliance Checker Skill (组合技能)

这是一个专为电商素材设计的、包含两个核心阶段的组合技能，旨在提供快速且深入的合规风险排查。它现在明确包含一个用于视频的专用审核流程。

子技能

• video-check: 此技能是本工作流的一部分，用于执行视频内容的初步API审查。

核心流程

工作流 1: 图片素材审核

1. 第一阶段：API 快速筛查 (调用 api-check 技能)

   • 目的：快速过滤掉明确、通用的违规内容。
   • 执行：调用 api-check 技能，对素材中的所有文案进行基础风险扫描，并对图片进行图像审核。
   • 审查范围：色情、暴恐、政治敏感、违禁品、辱骂等。
   • 结果：如果发现高风险内容，流程可以提前终止并要求修改。

2. 第二阶段：AI 广告法深度分析 (逻辑步骤)

   • 目的：在通过第一阶段后，专门审查特定于广告场景的合规风险。
   • 执行：AI读取 prompt_template.txt，并将知识库 (knowledge_base/) 和待审内容填充进去。
   • 审查重点：功效宣传、绝对化用语、违规承诺、误导性背书、画面元素风险等。
   • 输出：生成一份包含风险点和具体修改建议的详细报告。

工作流 2: 视频素材审核

1. 第一步：百度短视频审核 (调用 video-check 技能)

   • 目的：利用百度API对视频内容进行初步的、自动化的合规审查。
   • 执行：调用 video-check 子技能，提交视频文件或URL进行分析。

2. 第二步：AI电商素材深度审核 (逻辑步骤，严格按照步骤)

   • 目的：针对电商视频的常见违规点进行深度的人工智能审查。
   • 执行：
     1. AI读取并使用提示词 ecommerce_prompt.txt （在video-check技能目录下）作为其审查指令。
     2. AI将根据提示词中列出的规则，对视频内容进行逐一排查。
   • 审查重点:
     • 内容是否为单图/多图轮播
     • 是否存在低质量AI元素 (内容一眼假，人物不流畅，背景不真)
     • 是否存在过度P图，导致视频图片不符合常理
     • 视频内容是否呈现出截屏录屏特征
     • 是否包含封建迷信元素 (如出现钱币)

使用方法

这是一个逻辑组合技能，AI应根据素材类型选择相应的工作流，并按顺序调用所需的能力。

1. 接收素材 (图片/视频/文本)。
2. 选择工作流:
   • 如果素材是视频:
     1. (调用技能) 执行 video-check 进行初审。
     2. (逻辑步骤) 如果初审通过，则读取 ecommerce_prompt.txt 并执行AI二审。
   • 如果素材是文本或图片:
     1. (调用技能) 执行 api-check 进行快筛。
     2. (逻辑步骤) 如果快筛通过，则读取 prompt_template.txt 和 knowledge_base/ 的内容，执行AI深度分析。
3. 生成最终报告，汇总所有审查阶段的结果和修改建议。