短视频发布策略生成器

根据IP背景和视频文案，为短视频生成一套完整的发布策略，包括发布时间、话题标签、封面文案和发布标题，并能将文案合成为封面图。

Security Scan

The skill's files and runtime code broadly match the described purpose (strategy + cover image generation) but include unclear credential access and logging that are not declared in the metadata, which is inconsistent and warrants caution.

Purpose & Capability

The description (generate publishing strategy and compose cover images) matches the included prompts and image-generation script. However, the image script initializes an OpenAI client and retrieves an API key via get_config_value('qdd') even though the skill metadata declares no credentials or environment variables. Requesting an API key (or platform config) is expected for image generation, but it should be declared explicitly. This mismatch is incoherent.

Instruction Scope

SKILL.md instructs reading local prompt files and running the provided Python script for cover generation (expected). However, the PowerShell example uses hard-coded absolute paths and a Windows-specific Python path, which is brittle. More importantly, the generate_cover_image.py will call an external image-generation API and writes the full API response to api_response_content.log; that log may contain sensitive data. The first-step guidance forbids executing external scripts, yet run.py constructs a sessions_spawn call which implies spawning a sub-agent — the division of responsibilities is somewhat ambiguous.

Install Mechanism

There is no install spec (instruction-only plus included scripts). No external installers or downloads are invoked by the package metadata itself, which keeps install risk low. The included Python scripts will be executed if the user runs them, but nothing in the registry installation writes or executes arbitrary external code automatically.

Credentials

The code attempts to obtain an API key via get_config_value('qdd') and also requires a --base-url and --model when running the script, but the skill metadata declares no required env vars or primary credential. That is a clear mismatch: the runtime needs credentials/config but the skill does not declare them. Additionally, the script logs full API responses to disk, which increases the chance of exposing secrets or confidential content. This is disproportionate to the high-level purpose unless the author explicitly requests and documents the needed credential(s).

Persistence & Privilege

The skill does not request always:true and does not modify other skills. It writes output files and an api_response_content.log in its working directory (normal for debugging but could persist sensitive responses). run.py builds a sessions_spawn call (to spawn a sub-agent) — spawning sub-agents is an expected pattern here but grants more autonomous actions; combine that with the undeclared credential access for higher risk.

What to consider before installing

This skill appears to implement the advertised features, but the code expects an API key/config (get_config_value('qdd')) and uses an external image-generation API while the skill metadata declares no required credentials — that's inconsistent. Before installing or running: (1) ask the author to declare exactly which credentials or config keys are needed and why (e.g., OPENAI_API_KEY or a platform-specific key), (2) inspect or replace the undefined get_config_value function to ensure it does not fetch secrets from unexpected places, (3) be aware the script writes full API responses to api_response_content.log (remove or sanitize this logging if you care about confidentiality), (4) run the scripts in an isolated environment (no sensitive credentials) first to observe behavior, and (5) request cross-platform path fixes (avoid hard-coded Windows paths) and clearer instructions about when the agent should spawn sub-agents. If you cannot get satisfactory answers from the author, treat the skill as potentially risky and avoid providing platform-wide secrets to it.

Video Publisher Pro Skill

这是一个包含两个步骤的组合技能，旨在为短视频提供从策略到视觉呈现的完整发布方案。

---

流程

1. 第一步：策略生成 (逻辑技能)
   • 通过分析IP背景、周计划、视频文案和定位，生成专业的发布策略，包括发布时间、话题标签、封面文案和发布标题。
2. 第二步：封面生成 (脚本技能)
   • 将第一步产出的封面文案，通过AI图像生成模型，添加到用户指定的背景图片上，生成最终的视频封面。

---

第一步：策略生成

核心功能

通过深度分析用户提供的多维度信息，自动化地生成一套包含四大核心模块的发布建议。

使用方法 (重要)

这是一个纯粹的逻辑技能，其核心是通过一个精心设计的Prompt来调用AI自身的语言模型能力。

AI不应尝试执行任何外部脚本来完成此步骤。

正确的流程是：

1. 读取 prompt_template.txt 文件的内容。
2. 将用户提供的四类信息填充到模板中。
3. 将填充完毕的完整Prompt作为输入，调用语言模型，生成最终的策略报告。

所需信息

1. 客户背景信息: IP定位、核心价值观、目标用户画像等。
2. 本周计划或战略目标: 视频在整体策略中的位置和作用。
3. 视频文案: 完整的视频文案内容。
4. 视频定位: 视频的发布序列和核心目的（如：人设定位、干货教学等）。

---

第二步：封面生成

核心功能

使用 generate_cover_image.py 脚本，将文字标题智能地添加到背景图上。该技能现在内置一个自动判断机制，以确保输出结果的稳定性和可预测性。

核心工作流 (必须严格遵守)

1. 前置分析与模式选择 (AI视觉判断)

在执行任何生成操作前，必须先对用户提供的背景图进行AI视觉分析，判断其属于以下哪种模式：

• 版式复刻模式 (Replication Mode):

  • 触发条件: 图片上已存在清晰的、可作为参考的文字版式。
  • 核心目标: 只修改文字内容，不改变版式和风格。

• 智能融合模式 (Fusion Mode):

  • 触发条件: 图片上无明显文字，是一张干净的背景图。
  • 核心目标: 在不破坏画面的前提下，创作出与图片风格和谐的文字设计。

2. 根据模式调用相应指令

版式复刻模式 (Replication Mode)

当判断为“版式复刻”时，必须读取 replication_prompt.txt 文件的内容作为指令。

• 指令核心:

  1. 识别并锁定 原图中所有文字区块。
  2. 分析 每个区块的精确位置、字体、颜色和风格。
  3. 移除 旧文字。
  4. 将新文案按顺序填充回各自原来的位置和样式中。
  5. 输出图片。

• 最终效果: 实现对原始版式的精确“像素级”复刻，确保高效和稳定。

智能融合模式 (Fusion Mode)

当判断为“智能融合”时，必须读取 fusion_prompt.txt 文件的内容作为指令。

• 指令核心:

  1. 分析 图片的构图、视觉焦点和色彩氛围。
  2. 设计 与图片风格协调的文字样式（字体、颜色）。
  3. 在最佳位置（不遮挡主体）放置文案。
  4. 输出图片。

• 最终效果: 生成兼具设计感和信息清晰度的优质封面。

PowerShell 执行模板

# 1. 判断模式 (AI逻辑判断)
# $mode = "Replication" or "Fusion"

# 2. 根据模式选择Prompt文件路径
$prompt_file_path = if ($mode -eq "Replication") { "C:\\...\\replication_prompt.txt" } else { "C:\\...\\fusion_prompt.txt" }

# 3. 读取并填充Prompt
$prompt_template = Get-Content -Path $prompt_file_path -Raw
$final_prompt = $prompt_template -replace '\{text_to_add\}', $text_to_add

# 4. 执行Python脚本
& "F:\\python 3.10\\python.exe" "C:\\...\\generate_cover_image.py" --image $image_path --text $text_to_add --output $output_path --prompt $final_prompt

此升级旨在将该技能从一个简单的工具，转变为一个可靠、智能的自动化解决方案，杜绝“创意抽卡”，确保交付结果的确定性。