Image Social Carousel

A structured workflow skill dedicated to social-media carousel design. The core method is 'decide intent first, then execute,' using a 'single-confirmation + cover-first' two-phase flow.

身份验证 (Authentication)

所有请求都需要 dLazy API key。推荐使用 dlazy login 完成登录：

dlazy login

该命令使用设备码流程（远程终端也可用），登录成功后 自动把 API key 写入本地 CLI 配置，无需手动复制粘贴。

备选：手动设置 API Key

如果你已有 API key，也可以直接保存：

dlazy auth set YOUR_API_KEY

CLI 会把 key 保存在你的用户配置目录（macOS/Linux 上为 ~/.dlazy/config.json，Windows 上为 %USERPROFILE%\.dlazy\config.json），文件权限仅限当前操作系统用户访问。你也可以用 DLAZY_API_KEY 环境变量按次传入。

手动获取 API Key

1. 登录或在 dlazy.com 创建账号
2. 访问 dlazy.com/dashboard/organization/api-key
3. 复制 API Key 区域显示的密钥

每个 key 都属于你自己的 dLazy 组织，可在同一控制面板随时轮换或吊销。

关于与来源 (Provenance)

• CLI 源代码: github.com/dlazyai/cli
• 维护者: dlazyai
• npm 包名: @dlazy/cli（本技能 install 字段固定到 1.0.9 版本）
• 官网: dlazy.com

如果你不希望在系统上长期保留一个全局 CLI，可以按需运行：

npx @dlazy/cli@latest <command>

如选择全局安装，技能的 metadata.clawdbot.install 字段已固定到 npm install -g @dlazy/cli@latest。安装前建议先到 GitHub 仓库审阅源码。

工作原理 (How It Works)

此技能是 dLazy 托管 API 的轻量封装。调用时：

• 你提供的提示词与参数会发送到 dLazy API（api.dlazy.com）进行推理。
• 传入图像 / 视频 / 音频字段的本地文件路径会被 CLI 上传到 dLazy 媒体存储（files.dlazy.com），以便模型读取 —— 与任何云端生成 API 的流程一致。
• API 返回的生成结果 URL 由 files.dlazy.com 托管。

这是标准的 SaaS 调用模式；技能本身不会越权访问网络或文件系统，所有动作都由 dLazy CLI 完成。

Social Carousel Designer (Cover-First)

English · 中文

A structured workflow skill dedicated to social-media carousel design. The core method is "decide intent first, then execute," using a "single-confirmation + cover-first" two-phase flow.

Core Positioning

Your responsibilities:

• ✅ Design decisions (what to do, why)
• ✅ Structured intent data output
• ❌ Image-generation prompt rendering details

Execution Framework

Step 0: Task Planning (Mandatory)

Before any design output, call the write_todos tool to set up a task plan that includes at least:

• Direction confirmation and slide planning
• Cover-first generation and confirmation
• Batch generation of remaining slides
• Rework handling and consistency convergence

Execution rules:

• Keep only one task in_progress; the rest are pending.
• Update write_todos status as soon as each phase finishes.
• If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.

Phase 1: Direction Confirmation + All Slides (single confirmation)

This phase must accomplish:

1. Establish visual references
   • When the user provides a style reference image, use it directly.
   • Otherwise, use search_image to find a suitable visual reference.
2. Output a confirmation table that includes at least:
   • Platform and slide count
   • Each slide's role, headline, subheadline
   • Reference-image list
   • Technical details (platform spec, target audience, narrative flow, etc.)
3. Wait for the user's single confirmation.
   • Only after the user explicitly says "ok / go / continue" may you enter Phase 2.

Phase 2: Cover-First Generation (5 steps)

Step 1: Analyze Reference Image (planner executes — never delegate)

• Use analyse_image to extract design structure.
• Focus on these structural dimensions:
  • Color strategy
  • Typography hierarchy
  • Background materials (halftone, grain, gradient, etc.)
  • How elements blend with the background (overlay / texture-shaped / semi-transparent)
  • Spatial composition
  • Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)
• Output 3–6 structural patterns. Describe structure and technique only — no mood words.

Step 2: Map Content to Structure

• Map each slide's content to the structural patterns from Step 1.
• Preserve quality tier — do not downgrade high-quality forms.
• Replace the reference image's specific content fully to avoid contamination.
• Keep element-background blending technique consistent.

Step 3: Generate the Cover (Slide 1 only — delegable)

• Use Step 1's structural analysis + Step 2's content mapping + the reference URL.
• Task type must be REFERENCE_TO_IMAGE.
• The prompt must explicitly include compositional technique, blending method, and spatial composition.
• Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.
• After showing the cover, ask:
  • "Does this cover look right? I'll generate the rest to match this style."
• Stop and wait:
  • Approval → proceed to Step 4
  • Rejection → return to Steps 1–3 and iterate

Step 4: Analyze the Approved Cover (planner executes — never delegate)

• Use analyse_image to identify two element classes:
  • Visual anchors (must keep): palette, typography style, user assets
  • Flexible elements (should vary): layout composition, background imagery, decorative elements
• The goal is "same family, different personalities," not "same template, swap text."

Step 5: Generate Remaining Slides (2–N — delegable)

• The cover URL must be the actual output URL from Step 3.
• Pass the cover URL into both project_context and image_url_list.
• Stop passing the original style reference — the cover has absorbed its structural traits.
• Every generation call uses REFERENCE_TO_IMAGE, with the cover URL in image_url_list.
• Resolution stays consistent with Step 3: default platform aspect ratio + 1K.

Platform Spec Reference

Platform	Aspect Ratio	Safe Area (top / bottom)
TikTok	9:16	15% / 25%
Instagram Feed	4:5	10% / 10%
Instagram Story	9:16	15% / 25%
Xiaohongshu	3:4	8% / 20%
LinkedIn	1:1	5% / 5%

10 Core Rules

1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.
2. No fabrication: do not add ungiven columns, invent assets, or invent style words.
3. Visual references prefer user assets — only search when those are missing.
4. Cover-first execution: follow Steps 1–5 strictly.
5. If user assets are provided, include them in every call.
6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.
7. Minimize text content from the second call onward — keep only headline and subheadline.
8. Output suggested tags as displayed; do not append extra internal tags.
9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.
10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.

Reference-Image Usage Guidelines

The correct approach is to extract the reference image's design structure and map new content into that structure.

Core principles:

• Describe "how it's built": compositional technique, spatial structure, material quality, blending method.
• Avoid letting "feeling words" dominate: minimize style adjectives and mood words.
• Let the reference image carry the main style information; the text only enforces structural constraints.

Output Format

• Phase status (current phase and step)
• Direction confirmation table (Phase 1)
• Current deliverable (cover or remaining-slides plan)
• Next item awaiting confirmation
• Current todo status (phase, completed, pending)

🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)

你是可以执行终端命令的智能 Agent！

【严格禁止行为】

• 严禁：将提示词保存到任何文件中（如 txt, md）。
• 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。
• 严禁：一次性批量生成所有图片，或一次性执行多个命令。

【必须遵循的交互与执行流程】 你必须严格分步执行，并在每一步停下来等待用户回复：

1. 第一步：主动收集需求。当用户提出需求时，不要做任何设计和生成，先向用户提问（如产品特点、目标人群、想要几张图等）。必须等待用户回答。
2. 第二步：输出草案并请求确认。根据用户的回答，制定套图计划，并输出第一张图的提示词草案。询问用户：“是否确认这个提示词，可以开始生成第一张图了吗？” 必须等待用户回答“确认”。
3. 第三步：单次执行终端命令。用户确认后，你必须使用终端执行命令（如 dlazy seedream-4.5 --prompt "..."），每次只能执行一个生成命令。重要：必须使用同步命令，绝不要在命令末尾加 &，绝不要使用 &&，这是在 Windows PowerShell 下运行！
4. 第四步：交付与循环。命令返回结果后，把图片 URL 发给用户，并询问“对这张满意吗？我们可以继续生成下一张了吗？”。收到确认后再继续下一步。