{"id":"cf006407-6c6b-4165-812f-67df96eb695c","entityType":"agent","slug":"clawhub-dlazyai-dlazy-image-social-carousel","name":"社交轮播图设计 Social Carousel","canonicalUrl":"https://www.xpersona.co/agent/clawhub-dlazyai-dlazy-image-social-carousel","canonicalPath":"/agent/clawhub-dlazyai-dlazy-image-social-carousel","generatedAt":"2026-10-09T20:48:06.215Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":null},"description":"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.","descriptionLabel":"Source description","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 4.7K downloads reported by the source. Last updated 10/9/2026.","installCommand":"clawhub skill install s170j1ymymrxasgd00dsk7tckx84cf45:dlazy-image-social-carousel","sourceUrl":"https://clawhub.ai/dlazyai/dlazy-image-social-carousel","homepage":"https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/dlazyai/dlazy-image-social-carousel","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":74,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"社交轮播图设计 Social Carousel technical dossier on Xpersona with agent coverage, OPENCLEW support, and live trust metadata."},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":null},"stars":null,"forks":null,"downloads":4745,"packageName":null,"latestVersion":"1.3.27","tractionLabel":"4.7K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":null},"lastUpdatedAt":"2026-10-09T04:55:58.625Z","lastCrawledAt":"2026-10-09T04:55:58.625Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-10T04:55:58.625Z","lastVerifiedAt":null,"highlights":[{"version":"1.3.27","createdAt":"2026-10-08T01:14:55.297Z","changelog":"例行版本更新 2026-10-08","fileCount":4,"zipByteSize":12513},{"version":"1.3.26","createdAt":"2026-10-04T01:13:43.865Z","changelog":"例行版本更新 2026-10-04","fileCount":4,"zipByteSize":12510},{"version":"1.3.25","createdAt":"2026-10-02T04:56:18.805Z","changelog":"例行版本更新 2026-10-02","fileCount":4,"zipByteSize":12482},{"version":"1.3.24","createdAt":"2026-09-30T01:12:39.891Z","changelog":"例行版本更新 2026-09-30","fileCount":4,"zipByteSize":12577},{"version":"1.3.23","createdAt":"2026-09-28T02:11:39.999Z","changelog":"例行版本更新 2026-09-28","fileCount":4,"zipByteSize":12566},{"version":"1.3.22","createdAt":"2026-09-28T01:13:39.593Z","changelog":"例行版本更新 2026-09-28","fileCount":4,"zipByteSize":12514},{"version":"1.3.21","createdAt":"2026-09-24T01:50:15.352Z","changelog":"例行版本更新 2026-09-24","fileCount":4,"zipByteSize":12668},{"version":"1.3.20","createdAt":"2026-09-22T01:11:19.934Z","changelog":"例行版本更新 2026-09-22","fileCount":4,"zipByteSize":12609}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s170j1ymymrxasgd00dsk7tckx84cf45:dlazy-image-social-carousel","setupComplexity":"low","setupSteps":["Install using `clawhub skill install s170j1ymymrxasgd00dsk7tckx84cf45:dlazy-image-social-carousel` in an isolated environment before connecting it to live workloads.","No published capability contract is available yet, so validate auth and request/response behavior manually.","Review the upstream CLAWHUB listing at https://clawhub.ai/dlazyai/dlazy-image-social-carousel before using production credentials."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-09T20:48:06.209Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-dlazyai-dlazy-image-social-carousel/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":null},"readme":"Skill: 社交轮播图设计 Social Carousel\n\nOwner: dlazyai\n\nSummary: 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.\n\nTags: latest:1.3.27\n\nVersion history:\n\nv1.3.27 | 2026-10-08T01:14:55.297Z | user\n\n例行版本更新 2026-10-08\n\nv1.3.26 | 2026-10-04T01:13:43.865Z | user\n\n例行版本更新 2026-10-04\n\nv1.3.25 | 2026-10-02T04:56:18.805Z | user\n\n例行版本更新 2026-10-02\n\nv1.3.24 | 2026-09-30T01:12:39.891Z | user\n\n例行版本更新 2026-09-30\n\nv1.3.23 | 2026-09-28T02:11:39.999Z | user\n\n例行版本更新 2026-09-28\n\nv1.3.22 | 2026-09-28T01:13:39.593Z | user\n\n例行版本更新 2026-09-28\n\nv1.3.21 | 2026-09-24T01:50:15.352Z | user\n\n例行版本更新 2026-09-24\n\nv1.3.20 | 2026-09-22T01:11:19.934Z | user\n\n例行版本更新 2026-09-22\n\nv1.3.19 | 2026-09-20T01:17:09.283Z | user\n\n例行版本更新 2026-09-20\n\nv1.3.18 | 2026-09-18T01:37:48.259Z | user\n\n例行版本更新 2026-09-18\n\nv1.3.17 | 2026-09-14T01:13:40.360Z | user\n\n例行版本更新 2026-09-14\n\nv1.3.16 | 2026-09-10T01:10:53.625Z | user\n\n例行版本更新 2026-09-10\n\nv1.3.15 | 2026-09-08T01:11:33.467Z | user\n\n例行版本更新 2026-09-08\n\nv1.3.14 | 2026-09-07T01:15:02.058Z | user\n\n例行版本更新 2026-09-07\n\nv1.3.13 | 2026-09-04T01:17:02.916Z | user\n\n例行版本更新 2026-09-04\n\nv1.3.12 | 2026-09-02T01:11:36.215Z | user\n\n例行版本更新 2026-09-02\n\nv1.3.11 | 2026-08-31T02:52:15.933Z | user\n\n源码仓库迁移至 github.com/dlazy-ai/cli\n\nv1.3.10 | 2026-08-24T09:23:08.041Z | user\n\nRestore Chinese display name\n\nv1.3.9 | 2026-08-17T09:05:17.840Z | user\n\nRestore Chinese display name\n\nv1.3.8 | 2026-08-10T05:46:52.766Z | user\n\nRestore Chinese display name\n\nv1.3.7 | 2026-08-10T04:01:25.108Z | user\n\nRestore Chinese display name\n\nv1.3.6 | 2026-07-21T02:18:04.013Z | auto\n\n- Updated dLazy CLI minimum required version to 1.2.3.\n- Adjusted install instructions and metadata to point to CLI version 1.2.3.\n- Removed the redundant skill-card.md file.\n- No changes to core workflow or functionality.\n\nv1.3.5 | 2026-07-20T05:53:04.669Z | user\n\n规范 dlazy- 前缀命名，补充中文显示名\n\nv1.3.4 | 2026-07-17T10:38:08.932Z | auto\n\n- Updated documentation in SKILL.md and SKILL-cn.md for improved clarity and instructions.\n- Removed the skill-card.md file to streamline project documentation.\n- Bumped internal version from 1.2.1 to 1.2.2.\n- No changes to workflow logic or APIs; update is documentation-only.\n\nv1.2.2 | 2026-07-17T08:43:59.447Z | auto\n\n- Updated documentation files (SKILL.md and SKILL-cn.md) with refinements and minor corrections.\n- Removed file: skill-card.md.\n- No changes to the skill logic or CLI/API usage.\n- Ensured metadata and workflow descriptions are consistent and clear.\n\nv1.3.3 | 2026-07-09T01:47:56.112Z | user\n\nset Chinese display name for CN search\n\nv1.3.2 | 2026-07-09T01:47:43.897Z | user\n\nset Chinese display name for CN search\n\nv1.3.1 | 2026-07-09T01:29:35.466Z | user\n\nset Chinese display name for CN search\n\nv1.3.0 | 2026-07-07T19:43:16.800Z | auto\n\n- Updated skill version to 1.3.0.\n- SKILL.md and SKILL-cn.md content refreshed.\n- No workflow, rule, or functionality changes detected; documentation only.\n- Ensured descriptions, rules, and references are current and consistent in both languages.\n\nv1.2.1 | 2026-07-07T18:32:39.497Z | auto\n\n- Added bilingual (English/Chinese) description to improve clarity and usability for both audiences.\n- Enhanced documentation with explanations specifically tailored for Xiaohongshu, Instagram, and similar platforms.\n- Removed redundant file skill-card.md to streamline resource management.\n- No workflow or technical changes; documentation now emphasizes \"decide intent first, execute later\" and the cover-first, two-phase approach.\n- Improved consistency and alignment between English and Chinese docs for easier referencing.\n\nv1.2.0 | 2026-06-02T10:04:14.832Z | auto\n\nimage-social-carousel v1.2.0\n\n- Updated CLI installation instructions: now recommends using the latest @dlazy/cli version instead of locking to 1.0.9.\n- Updated metadata fields to use @dlazy/cli@latest for both global and npx installs.\n- Removed the separate skill-card.md file.\n- Documentation clarifies use of @latest throughout, improving future compatibility and reducing maintenance.\n- No changes to workflow logic, execution phases, or design rules.\n\nv1.1.2 | 2026-06-02T01:40:41.304Z | user\n\nUpdate model skills (names/params refreshed); add search-audio/image/video\n\nv1.1.1 | 2026-05-06T14:13:39.565Z | auto\n\nimage-social-carousel 1.1.1\n\n- Documentation updates: SKILL.md and SKILL-cn.md content revised.\n- No functional or workflow changes; internal version bump and doc maintenance only.\n\nv1.1.0 | 2026-05-01T07:56:44.550Z | auto\n\n- Adds support for authentication via `dlazy login` (device code flow), simplifying API key management.\n- Keeps manual API key setting as an alternative, with clearer, updated instructions.\n- The authentication section now recommends logging in instead of manually copying API keys.\n- No changes to workflow or generation logic—update focuses on setup clarity and user onboarding.\n\nv1.0.9 | 2026-04-30T10:22:45.991Z | user\n\nSync to CLI 1.0.9 with cliWarning support\n\nv1.0.5 | 2026-04-29T02:35:52.037Z | user\n\nbump @dlazy/cli to 1.0.8\n\nv1.0.4 | 2026-04-27T08:03:55.741Z | user\n\nReduce false-positive scanner alerts: drop 'plaintext' wording from API key storage docs; remove persistsApiKey/network metadata flags in favour of neutral configLocation/apiEndpoints; rewrite Data & Privacy section as factual How-It-Works description without alarming warnings; emphasise that keys can be rotated/revoked at any time from the dLazy dashboard.\n\nv1.0.3 | 2026-04-27T07:42:39.023Z | user\n\nAdd provenance metadata (homepage/source/author/npm), document API key storage location (~/.dlazy/config.json) and DLAZY_API_KEY env var alternative, add Data & Privacy section, recommend 'npx @dlazy/cli@1.0.6' install alternative, normalise Chinese auth-error instruction wording.\n\nv1.0.2 | 2026-04-24T04:09:32.794Z | auto\n\n- Updated documentation in SKILL.md, SKILL-en.md, and SKILL-cn.md to clarify instructions and workflow.\n- No changes to core logic or code; this release focuses on improved clarity and guidance for users.\n\nv1.0.1 | 2026-04-24T03:35:57.318Z | auto\n\n- 新增“身份验证”说明，包括获取和设置 dLazy API Key 的步骤。\n- 更新 dlazy CLI 依赖版本到 1.0.6。\n- 文档补充 CLI 配置细节，提升集成指引清晰度。\n\nv1.0.0 | 2026-04-23T05:44:37.466Z | auto\n\nInitial release of a structured workflow skill for designing social media image carousels.\n\n- Introduces a two-phase process: design intent confirmation followed by generation, with a \"one-time confirmation + cover-first\" workflow.\n- Provides detailed step-by-step execution guidelines, including planning, visual reference selection, structured intent output, and iterative generation.\n- Enforces strict rules on image generation (no batch execution, user assets prioritized, specific platform specs).\n- Includes a mandatory confirmation checkpoint before image rendering, and a feedback-driven process for revisions and consistency.\n- Outlines precise usage instructions for terminal commands and output formats.\n\nArchive index:\n\nArchive v1.3.27: 4 files, 12513 bytes\n\nFiles: skill-card.md (1995b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.27:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.27:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.27\",\n  \"publishedAt\": 1791422095297\n}\n\nFile v1.3.27:skill-card.md\n\n## Description:\n\nGuides agents through planning and generating social-media carousel slides using a confirmed design direction and an approved cover image.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nContent creators and marketing teams use this skill to plan platform-specific social carousels, approve a cover design, and generate visually consistent remaining slides.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Using an external CLI package requires installing or running third-party software.\n\nMitigation: Review the @dlazy/cli package before use; choose the npx alternative if a global installation is undesirable.\n\nRisk: Prompts and referenced media are sent to dLazy's hosted service, and use requires an API key.\n\nMitigation: Avoid submitting sensitive content without approval and rotate or revoke the API key through the service dashboard when needed.\n\n## Reference(s):\n\n- [ClawHub skill listing](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI repository (skill metadata link)](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Guidance, Shell commands]\n\n**Output Format:** [Markdown planning tables, status updates, generation commands, and generated-image URLs]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Slide plans and image generation proceed after user confirmation; the cover is approved before remaining slides.]\n\n## Skill Version(s):\n\n1.3.27 (source: server-resolved release metadata)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.27:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.26: 4 files, 12510 bytes\n\nFiles: skill-card.md (1993b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.26:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.26:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.26\",\n  \"publishedAt\": 1791076423865\n}\n\nFile v1.3.26:skill-card.md\n\n## Description:\n\nGuides agents through planning, confirming, and creating social-media carousel images with a cover-first workflow.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nCreators and marketing teams use this skill to plan platform-specific social carousels, approve a visual direction and cover, then generate coordinated remaining slides through the dLazy CLI.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The referenced CLI package and version may differ from descriptions in the skill.\n\nMitigation: Verify the @dlazy/cli package and pinned version before installation.\n\nRisk: Prompts and referenced local media are sent to dLazy's cloud service.\n\nMitigation: Obtain approval before uploading sensitive prompts or media and review what is shared.\n\nRisk: The CLI may store an API key locally, and responses may include Chinese prompts.\n\nMitigation: Use DLAZY_API_KEY if local key storage is unsuitable, and review output language before use.\n\n## Reference(s):\n\n- [ClawHub skill listing](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI source](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Markdown, Shell commands, Image URLs]\n\n**Output Format:** [Markdown with planning tables, commands, and links to generated images]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Presents a direction plan and cover for approval before proceeding with remaining slides.]\n\n## Skill Version(s):\n\n1.3.26 (source: ClawHub release)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.26:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.25: 4 files, 12482 bytes\n\nFiles: skill-card.md (1861b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.25:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.25:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.25\",\n  \"publishedAt\": 1790916978805\n}\n\nFile v1.3.25:skill-card.md\n\n## Description:\n\nGuides social-media carousel design through slide planning, user confirmation, and cover-first image generation.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nCreators and marketing teams use this skill to plan platform-specific social carousels, approve a cover design, and generate consistent follow-on slides with dLazy.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Prompts and selected local media are sent to dLazy-hosted services.\n\nMitigation: Send only assets and content you are comfortable sharing with dLazy.\n\nRisk: Installing the CLI and storing a persistent API key expand local credential exposure.\n\nMitigation: Use the on-demand npx option and per-session DLAZY_API_KEY when appropriate; rotate or revoke keys when no longer needed.\n\n## Reference(s):\n\n- [ClawHub skill listing](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI source (listed in skill metadata)](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Markdown, Image URLs, Guidance]\n\n**Output Format:** [Markdown slide plans, status updates, and links to generated images]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Includes a direction-confirmation table and cover approval before remaining slides.]\n\n## Skill Version(s):\n\n1.3.25 (source: ClawHub release metadata; bundled frontmatter states 1.3.6)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.25:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.24: 4 files, 12577 bytes\n\nFiles: skill-card.md (2129b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.24:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.24:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.24\",\n  \"publishedAt\": 1790730759891\n}\n\nFile v1.3.24:skill-card.md\n\n## Description:\n\nGuides the design of social-media carousels through slide planning, user confirmation, and cover-first image generation.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nCreators and marketing teams use this skill to plan platform-specific social-media carousels, approve a visual direction and cover, and generate coordinated slides with dLazy image services.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Installing or running a third-party CLI executes software on the user's device.\n\nMitigation: Review the dLazy CLI source before choosing a global npm installation or npx execution.\n\nRisk: A dLazy API key is stored locally when signing in through the CLI.\n\nMitigation: Protect local credentials and rotate or revoke the key if it is exposed.\n\nRisk: Prompts and referenced media files are sent to dLazy-hosted services.\n\nMitigation: Obtain permission for supplied media and avoid submitting sensitive content unless the service is approved for it.\n\n## Reference(s):\n\n- [ClawHub skill release](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI repository (listed in skill metadata)](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Design guidance, Markdown, Shell commands, Image URLs]\n\n**Output Format:** [Markdown with slide plans, approval prompts, command snippets, and generated image links]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Requests user approval before image generation and checks the cover before continuing with other slides.]\n\n## Skill Version(s):\n\n1.3.24 (source: ClawHub release; supplied skill frontmatter lists 1.3.6)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.24:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.23: 4 files, 12566 bytes\n\nFiles: skill-card.md (2150b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.23:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.23:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.23\",\n  \"publishedAt\": 1790561499999\n}\n\nFile v1.3.23:skill-card.md\n\n## Description:\n\nGuides social-media carousel design through a single direction confirmation, cover approval, and generation of matching slides.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nCreators and marketers use this skill to plan platform-specific social carousels, approve a cover design, and generate a consistent set of slides through the dLazy CLI.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Prompts and referenced media may be sent to dLazy services, where generated images are hosted.\n\nMitigation: Avoid sending confidential content without approval; review dLazy's data handling before use.\n\nRisk: The dLazy API key may be stored in ~/.dlazy/config.json.\n\nMitigation: Use DLAZY_API_KEY per run when local storage is unsuitable, and rotate or revoke keys when needed.\n\nRisk: The bundled skill version differs from the ClawHub release version, and CLI installation instructions name a separate version.\n\nMitigation: Review the linked CLI package and source, and confirm the intended version before installation.\n\n## Reference(s):\n\n- [ClawHub skill release](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI source (skill metadata link; not verified release provenance)](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Markdown, Guidance, Shell commands, Image URLs]\n\n**Output Format:** [Markdown with slide plans, approval prompts, CLI commands, and links to generated images]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Images are generated and hosted by dLazy services.]\n\n## Skill Version(s):\n\n1.3.23 (source: ClawHub release; bundled skill frontmatter says 1.3.6)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.23:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.22: 4 files, 12514 bytes\n\nFiles: skill-card.md (2014b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.22:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.22:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.22\",\n  \"publishedAt\": 1790558019593\n}\n\nFile v1.3.22:skill-card.md\n\n## Description:\n\nGuides agents through planning and generating social-media carousel images using a cover-first approval workflow and the dLazy CLI.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nCreators and marketing teams use this skill to plan platform-specific social-media carousels, approve a cover image, and generate matching slides through dLazy.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Prompts and supplied media are sent to dLazy-hosted services.\n\nMitigation: Only submit content approved for third-party processing; review referenced files before use.\n\nRisk: Using the dLazy CLI requires an API key that may be persisted locally.\n\nMitigation: Use a per-session environment variable instead of saved credentials when persistence is undesirable.\n\nRisk: A global installation leaves the third-party CLI installed on the device.\n\nMitigation: Prefer the documented npx invocation when a global installation is unnecessary.\n\n## Reference(s):\n\n- [ClawHub skill release](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI source (tooling, not verified skill provenance)](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Guidance, Text, Shell commands]\n\n**Output Format:** [Markdown with planning tables, commands, and generated image URLs]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Cover-first review before generating remaining slides.]\n\n## Skill Version(s):\n\n1.3.22 (source: ClawHub release metadata; artifact frontmatter says 1.3.6)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.22:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.21: 4 files, 12668 bytes\n\nFiles: skill-card.md (2388b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.21:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nFile v1.3.21:_meta.json\n\n{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.21\",\n  \"publishedAt\": 1790214615352\n}\n\nFile v1.3.21:skill-card.md\n\n## Description:\n\nA 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.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nExternal users and content teams use this skill to plan social-media carousel image sets, confirm direction once, approve the cover first, and then generate matching remaining slides through the dLazy CLI workflow.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The workflow requires a dLazy API key and may store credentials in the local CLI configuration.\n\nMitigation: Use environment-based credentials when saved configuration is not desired, and rotate or revoke API keys from the dLazy dashboard when needed.\n\nRisk: Prompts and referenced media files may be sent to dLazy-hosted services for generation.\n\nMitigation: Avoid submitting sensitive prompts or media unless the user is comfortable with dLazy processing and hosting generated outputs.\n\nRisk: The skill depends on the @dlazy/cli package and may incur API usage costs.\n\nMitigation: Review the package and pinned version before installation, and confirm expected API usage costs before running generation commands.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI source](https://github.com/dlazy-ai/cli)\n- [@dlazy/cli npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, shell commands, configuration, guidance]\n\n**Output Format:** [Markdown with tables, status updates, prompts, CLI commands, and generated image URLs]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Uses user confirmation gates before image generation and sends prompts or referenced media to dLazy-hosted services when the CLI is run.]\n\n## Skill Version(s):\n\n1.3.21 (source: server release metadata; artifact frontmatter lists 1.3.6)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v1.3.21:SKILL-cn.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and consistency convergence\n\nExecution rules:\n\n- Keep only one task `in_progress`; the rest are `pending`.\n- Update `write_todos` status as soon as each phase finishes.\n- If the user asks for rework or new assets, add or re-order tasks and re-enter the corresponding phase.\n\n### Phase 1: Direction Confirmation + All Slides (single confirmation)\n\nThis phase must accomplish:\n\n1. Establish visual references\n   - When the user provides a style reference image, use it directly.\n   - Otherwise, use `search_image` to find a suitable visual reference.\n2. Output a confirmation table that includes at least:\n   - Platform and slide count\n   - Each slide's role, headline, subheadline\n   - Reference-image list\n   - Technical details (platform spec, target audience, narrative flow, etc.)\n3. Wait for the user's single confirmation.\n   - Only after the user explicitly says \"ok / go / continue\" may you enter Phase 2.\n\n### Phase 2: Cover-First Generation (5 steps)\n\n#### Step 1: Analyze Reference Image (planner executes — never delegate)\n\n- Use `analyse_image` to extract design structure.\n- Focus on these structural dimensions:\n  - Color strategy\n  - Typography hierarchy\n  - Background materials (halftone, grain, gradient, etc.)\n  - How elements blend with the background (overlay / texture-shaped / semi-transparent)\n  - Spatial composition\n  - Texture quality of key elements (photoreal 3D, flat vector, sculptural, etc.)\n- Output 3–6 structural patterns. Describe structure and technique only — no mood words.\n\n#### Step 2: Map Content to Structure\n\n- Map each slide's content to the structural patterns from Step 1.\n- Preserve quality tier — do not downgrade high-quality forms.\n- Replace the reference image's specific content fully to avoid contamination.\n- Keep element-background blending technique consistent.\n\n#### Step 3: Generate the Cover (Slide 1 only — delegable)\n\n- Use Step 1's structural analysis + Step 2's content mapping + the reference URL.\n- Task type must be `REFERENCE_TO_IMAGE`.\n- The prompt must explicitly include compositional technique, blending method, and spatial composition.\n- Default resolution: platform aspect ratio + 1K; only escalate when the user explicitly asks for more.\n- After showing the cover, ask:\n  - \"Does this cover look right? I'll generate the rest to match this style.\"\n- Stop and wait:\n  - Approval → proceed to Step 4\n  - Rejection → return to Steps 1–3 and iterate\n\n#### Step 4: Analyze the Approved Cover (planner executes — never delegate)\n\n- Use `analyse_image` to identify two element classes:\n  - Visual anchors (must keep): palette, typography style, user assets\n  - Flexible elements (should vary): layout composition, background imagery, decorative elements\n- The goal is \"same family, different personalities,\" not \"same template, swap text.\"\n\n#### Step 5: Generate Remaining Slides (2–N — delegable)\n\n- The cover URL must be the actual output URL from Step 3.\n- Pass the cover URL into both `project_context` and `image_url_list`.\n- Stop passing the original style reference — the cover has absorbed its structural traits.\n- Every generation call uses `REFERENCE_TO_IMAGE`, with the cover URL in `image_url_list`.\n- Resolution stays consistent with Step 3: default platform aspect ratio + 1K.\n\n## Platform Spec Reference\n\n| Platform | Aspect Ratio | Safe Area (top / bottom) |\n| --- | --- | --- |\n| TikTok | 9:16 | 15% / 25% |\n| Instagram Feed | 4:5 | 10% / 10% |\n| Instagram Story | 9:16 | 15% / 25% |\n| Xiaohongshu | 3:4 | 8% / 20% |\n| LinkedIn | 1:1 | 5% / 5% |\n\n## 10 Core Rules\n\n1. Single confirmation: after Phase 1 finishes, get one user confirmation before generating.\n2. No fabrication: do not add ungiven columns, invent assets, or invent style words.\n3. Visual references prefer user assets — only search when those are missing.\n4. Cover-first execution: follow Steps 1–5 strictly.\n5. If user assets are provided, include them in every call.\n6. Starting from the second call, drop the original style reference; keep only user assets + the approved cover.\n7. Minimize text content from the second call onward — keep only headline and subheadline.\n8. Output suggested tags as displayed; do not append extra internal tags.\n9. Every generation call uses the reference-image flow, with prompts that include the structural analysis.\n10. Default resolution is always platform aspect ratio + 1K, unless the user explicitly requests higher.\n\n## Reference-Image Usage Guidelines\n\nThe correct approach is to extract the reference image's design structure and map new content into that structure.\n\nCore principles:\n\n- Describe \"how it's built\": compositional technique, spatial structure, material quality, blending method.\n- Avoid letting \"feeling words\" dominate: minimize style adjectives and mood words.\n- Let the reference image carry the main style information; the text only enforces structural constraints.\n\n## Output Format\n\n- Phase status (current phase and step)\n- Direction confirmation table (Phase 1)\n- Current deliverable (cover or remaining-slides plan)\n- Next item awaiting confirmation\n- Current todo status (phase, completed, pending)\n\n\n## 🛠️ 执行与生成指南 (CRITICAL EXECUTION INSTRUCTIONS)\n\n**你是可以执行终端命令的智能 Agent！**\n\n**【严格禁止行为】**\n- 严禁：将提示词保存到任何文件中（如 txt, md）。\n- 严禁：要求用户自己去第三方平台（如 Midjourney）生成图片。\n- 严禁：一次性批量生成所有图片，或一次性执行多个命令。\n\n**【必须遵循的交互与执行流程】**\n你必须**严格分步**执行，并在每一步停下来等待用户回复：\n\n1. **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。**\n2. **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”。**\n3. **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！**\n4. **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。\n\nArchive v1.3.20: 4 files, 12609 bytes\n\nFiles: skill-card.md (2306b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nFile v1.3.20:SKILL.md\n\n---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dla\n\nArchive v1.3.19: 4 files, 12591 bytes\n\nFiles: skill-card.md (2126b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)\n\nArchive v1.3.18: 4 files, 12573 bytes\n\nFiles: skill-card.md (2172b), SKILL-cn.md (10719b), SKILL.md (10719b), _meta.json (147b)","readmeExcerpt":"Skill: 社交轮播图设计 Social Carousel Owner: dlazyai Summary: 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. Tags: latest:1.3.27 Version history: v1.3.27 | 2026-10-08T01:14:55.297Z | user 例行版本更新 2026-10-08 v1.3.26 | 2026-10-04T01:13:43.865Z | user 例行版本更新 2026-10-04 v1.3.25 | 2026-10-02T","codeSnippets":[],"executableExamples":[{"language":"bash","snippet":"dlazy login"},{"language":"bash","snippet":"dlazy auth set YOUR_API_KEY"},{"language":"bash","snippet":"npx @dlazy/cli@1.2.3 <command>"},{"language":"bash","snippet":"dlazy login"},{"language":"bash","snippet":"dlazy auth set YOUR_API_KEY"},{"language":"bash","snippet":"npx @dlazy/cli@1.2.3 <command>"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and con"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn7c5wgeajfcfvdfb5ceemvdb984cjpd\",\n  \"slug\": \"dlazy-image-social-carousel\",\n  \"version\": \"1.3.27\",\n  \"publishedAt\": 1791422095297\n}"},{"path":"skill-card.md","content":"## Description:\n\nGuides agents through planning and generating social-media carousel slides using a confirmed design direction and an approved cover image.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[dlazyai](https://clawhub.ai/user/dlazyai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nContent creators and marketing teams use this skill to plan platform-specific social carousels, approve a cover design, and generate visually consistent remaining slides.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Using an external CLI package requires installing or running third-party software.\n\nMitigation: Review the @dlazy/cli package before use; choose the npx alternative if a global installation is undesirable.\n\nRisk: Prompts and referenced media are sent to dLazy's hosted service, and use requires an API key.\n\nMitigation: Avoid submitting sensitive content without approval and rotate or revoke the API key through the service dashboard when needed.\n\n## Reference(s):\n\n- [ClawHub skill listing](https://clawhub.ai/dlazyai/skills/dlazy-image-social-carousel)\n- [dLazy CLI repository (skill metadata link)](https://github.com/dlazy-ai/cli)\n- [dLazy CLI npm package](https://www.npmjs.com/package/@dlazy/cli)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Guidance, Shell commands]\n\n**Output Format:** [Markdown planning tables, status updates, generation commands, and generated-image URLs]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Slide plans and image generation proceed after user confirmation; the cover is approved before remaining slides.]\n\n## Skill Version(s):\n\n1.3.27 (source: server-resolved release metadata)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment."},{"path":"SKILL-cn.md","content":"---\nname: dlazy-image-social-carousel\nversion: 1.3.6\ndescription: 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.\ntriggers:\n  - Social Carousel Designer (Cover-First)\nmetadata: {\"clawdbot\":{\"emoji\":\"🤖\",\"requires\":{\"bins\":[\"npm\",\"npx\"]},\"install\":\"npm install -g @dlazy/cli@1.2.3\",\"installAlternative\":\"npx @dlazy/cli@1.2.3\",\"homepage\":\"https://github.com/dlazy-ai/cli\",\"source\":\"https://github.com/dlazy-ai/cli\",\"author\":\"dlazyai\",\"license\":\"see-repo\",\"npm\":\"https://www.npmjs.com/package/@dlazy/cli\",\"configLocation\":\"~/.dlazy/config.json\",\"apiEndpoints\":[\"api.dlazy.com\",\"files.dlazy.com\"]},\"openclaw\":{\"systemPrompt\":\"当你需要使用此技能时，请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型（如 dlazy seedream-4.5 等）来完成实际的图片渲染。注意：Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行，请单独且同步地执行命令。\"}}\n---\n\n## 身份验证 (Authentication)\n\n所有请求都需要 dLazy API key。**推荐使用** `dlazy login` 完成登录：\n\n```bash\ndlazy login\n```\n\n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。\n\n### 备选：手动设置 API Key\n\n如果你已有 API key，也可以直接保存：\n\n```bash\ndlazy auth set YOUR_API_KEY\n```\n\nCLI 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。\n\n### 手动获取 API Key\n\n1. 登录或在 [dlazy.com](https://dlazy.com) 创建账号\n2. 访问 [dlazy.com/dashboard/organization/api-key](https://dlazy.com/dashboard/organization/api-key)\n3. 复制 API Key 区域显示的密钥\n\n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**。\n\n## 关于与来源 (Provenance)\n\n- **CLI 源代码**: [github.com/dlazy-ai/cli](https://github.com/dlazy-ai/cli)\n- **维护者**: dlazyai\n- **npm 包名**: `@dlazy/cli`（本技能 install 字段固定到 `1.0.9` 版本）\n- **官网**: [dlazy.com](https://dlazy.com)\n\n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：\n\n```bash\nnpx @dlazy/cli@1.2.3 <command>\n```\n\n如选择全局安装，技能的 `metadata.clawdbot.install` 字段已固定到 `npm install -g @dlazy/cli@1.2.3`。安装前建议先到 GitHub 仓库审阅源码。\n\n## 工作原理 (How It Works)\n\n此技能是 dLazy 托管 API 的轻量封装。调用时：\n\n- 你提供的提示词与参数会发送到 dLazy API（`api.dlazy.com`）进行推理。\n- 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。\n- API 返回的生成结果 URL 由 `files.dlazy.com` 托管。\n\n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。\n\n# 社交轮播图设计 Social Carousel\n\n[English](./SKILL.md) · [中文](./SKILL-cn.md)\n\nA 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.\n\n## Core Positioning\n\nYour responsibilities:\n\n- ✅ Design decisions (what to do, why)\n- ✅ Structured intent data output\n- ❌ Image-generation prompt rendering details\n\n## Execution Framework\n\n### Step 0: Task Planning (Mandatory)\n\nBefore any design output, call the `write_todos` tool to set up a task plan that includes at least:\n\n- Direction confirmation and slide planning\n- Cover-first generation and confirmation\n- Batch generation of remaining slides\n- Rework handling and con"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":null,"editorialQuality":{"score":100,"threshold":65,"status":"thin","wordCount":1248,"uniquenessScore":39,"reasons":["uniqueness-below-45"]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-09T04:55:58.625Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-09T20:48:06.215Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}