Hopola

Purpose

Hopola turns "web research → image/video/3D generation → result upload → report output" into a reusable workflow, supporting both one-shot end-to-end execution and stage-based execution.

When to Use

• The user needs content production based on web information and also needs visual outputs and delivery.
• The user wants one-pass execution for research, generation, upload, and consolidated reporting.
• The user only needs a specific stage (search only / generation only / upload only / report only).
• The user needs video or 3D outputs included in the same report.

Prerequisites

• web-access is available for web search and page reading.
• OpenClaw Gateway is reachable.
• OPENCLOW_KEY is configured in the runtime environment.
• Upload policy endpoint uses MAAT_TOKEN_API (compatible with MEITU_TOKEN_API / NEXT_PUBLIC_MAAT_TOKEN_API / NEXT_PUBLIC_MEITU_TOKEN_API).
• MAAT_TOKEN_API_ALLOWED_HOSTS controls trusted upload policy hosts (comma-separated, includes strategy.stariidata.com by default).
• OPENCLAW_REQUEST_LOG defaults to 0 and should be enabled only for temporary debugging.
• If OPENCLOW_KEY is missing, stop at precheck and return structured_error.code=GATEWAY_KEY_MISSING; do not call Gateway tools.

ClawHub Release Structure

• SKILL.md: Main orchestration and strategy
• subskills/search/SKILL.md: Search subskill
• subskills/generate-image/SKILL.md: Image generation subskill
• subskills/generate-video/SKILL.md: Video generation subskill
• subskills/generate-3d/SKILL.md: 3D generation subskill
• subskills/logo-design/SKILL.md: Logo design subskill
• subskills/product-image/SKILL.md: Product image subskill
• subskills/cutout/SKILL.md: Product cutout subskill
• subskills/upload/SKILL.md: Upload subskill
• subskills/report/SKILL.md: Report subskill
• playbooks/design-intake.md: Design intake template and proposal framework
• scripts/: Release validation and packaging scripts
• assets/: Logo, cover, and flow diagram
• README.zh-CN.md / README.en.md: Bilingual documentation

Execution Modes

Mode A: Single-Entry Full Pipeline

Execute in order:

1. Intake & Normalize: Collect explicit URLs plus session-uploaded images and normalize them into upload-ready assets.
2. Search: Invoke the search subskill to extract usable facts.
3. Discover: Query Gateway MCP tools and identify image/video/3D tools.
4. Generate: Invoke the corresponding generation subskill by task type.
5. Upload: Invoke the upload subskill to submit images or result links.
6. Report: Invoke the report subskill to output a Markdown report.

Mode B: Stage-Based Execution

Execute only one stage based on user instruction:

• SearchOnly
• GenerateImageOnly
• GenerateVideoOnly
• Generate3DOnly
• GenerateLogoOnly
• GenerateProductImageOnly
• CutoutOnly
• UploadOnly
• ReportOnly

Standard Inputs

• query: Search topic or target question.
• search_scope: Search scope and constraints (optional).
• image_prompt: Prompt for image generation (optional, required in GenerateImageOnly).
• video_prompt: Prompt for video generation (optional, required in GenerateVideoOnly).
• model3d_prompt: Prompt for 3D generation (optional, required in Generate3DOnly).
• logo_prompt: Prompt for logo generation (optional, required in GenerateLogoOnly).
• product_prompt: Prompt for product image generation (optional, required in GenerateProductImageOnly).
• product_image_url: Source product image URL (optional in input, mandatory before product-image generation tool call).
• image_url: Source image URL for cutout (optional, required in CutoutOnly).
• session_uploaded_images: Session-level uploaded image references (optional; local file path, attachment handle, or markdown image source).
• auto_upload_session_images: Whether to auto-upload session-uploaded images first (default true).
• normalized_source_images: Internal normalized image list produced by intake stage (optional internal field).
• primary_source_image_url: First available uploaded URL after normalization (optional internal field).
• cutout_mode: Cutout mode for edge handling (optional, default auto).
• output_background: Background mode after cutout (optional, default transparent).
• desired_scenes: Requested product image scenes list (optional, required before generation).
• product_info_confirmed: Whether product information is confirmed before generation (optional, default false).
• source_image_confirmed: Whether the source product image is explicitly provided or confirmed by user (optional, default false; mandatory before product-image generation).
• image_style: Style preference (optional).
• image_size: Image size (optional).
• video_ratio: Video aspect ratio (optional).
• video_duration: Video duration (optional).
• model3d_quality: 3D quality level (optional).
• upload_enabled: Whether to upload results (default true).
• report_format: Fixed as markdown.
• mode: pipeline or stage.
• task_type: image|video|3d|logo|product-image.
• stage: When mode=stage, use one of search|generate-image|generate-video|generate-3d|generate-logo|generate-product-image|cutout|upload|report.
• gateway_base_url: Default https://hopola.ai.
• openclaw_key_env: Default OPENCLOW_KEY.
• response_language: Optional explicit output language code (en, zh-CN, ja, etc.); if omitted, infer from the latest user message.

Standard Output (Markdown)

The report must include at least:

1. Summary of search findings
2. Generation outputs (image/video/3D)
3. Upload results
4. Security status and error alerts
5. Final conclusion and next-step suggestions

Configuration Placeholders

• GATEWAY_BASE_URL: Gateway domain
• GATEWAY_MCP_TOOLS_ENDPOINT: /api/gateway/mcp/tools
• GATEWAY_MCP_CALL_ENDPOINT: /api/gateway/mcp/call
• GATEWAY_KEY_HEADER: X-OpenClaw-Key
• MAAT_UPLOAD_SCRIPT_PATH: scripts/maat_upload.py
• IMAGE_PREFERRED_TOOL_NAME: Preferred fixed tool for image generation
• VIDEO_PREFERRED_TOOL_NAME: Preferred fixed tool for video generation
• MODEL3D_PREFERRED_TOOL_NAME: Preferred fixed tool for 3D generation
• MODEL3D_SECONDARY_PREFERRED_TOOL_NAME: Secondary preferred tool for 3D generation
• LOGO_PREFERRED_TOOL_NAME: Preferred fixed tool for logo generation
• PRODUCT_IMAGE_PREFERRED_TOOL_NAME: Preferred fixed tool for product image generation
• IMAGE_MCP_TOOL_MATCH_RULES: Auto-discovery rules for image tools
• VIDEO_MCP_TOOL_MATCH_RULES: Auto-discovery rules for video tools
• MODEL3D_MCP_TOOL_MATCH_RULES: Auto-discovery rules for 3D tools
• IMAGE_MCP_ARGS_SCHEMA: Argument mapping schema for image tools
• VIDEO_MCP_ARGS_SCHEMA: Argument mapping schema for video tools
• MODEL3D_MCP_ARGS_SCHEMA: Argument mapping schema for 3D tools
• LOGO_MCP_ARGS_SCHEMA: Argument mapping schema for logo tools
• PRODUCT_IMAGE_MCP_ARGS_SCHEMA: Argument mapping schema for product image tools
• REQUEST_TIMEOUT_MS: Request timeout
• RETRY_MAX: Max retries
• MAAT_TOKEN_API: Upload policy endpoint (default https://strategy.stariidata.com/upload/policy)
• MAAT_TOKEN_API_ALLOWED_HOSTS: Trusted hosts for upload policy endpoint
• OPENCLAW_REQUEST_LOG: Request log switch (0 by default)

Execution Rules

• In the search stage, prioritize strategic online retrieval with web-access; combine search and page reading when necessary.
• Before routing, normalize natural-language intent first: if user asks for product visuals such as 商品图/主图/详情图/场景图 or equivalent expressions, force task_type=product-image and stage=generate-product-image.
• For task_type=product-image or stage=generate-product-image, if no usable source image is present (product_image_url missing and session_uploaded_images empty/unresolvable), only return inquiry guidance asking user to upload product image or provide public URL; do not call any product-image generation tool.
• Before any stage requiring image URL input (product_image_url, image_url, model3d_image_url), run intake normalization:
  • Merge explicit URL fields with session_uploaded_images.
  • If an item is a local/session file, attachment reference, or markdown image source, invoke subskills/upload/SKILL.md to upload first.
  • Write uploaded links back to corresponding URL fields; if multiple links exist, use the first as primary_source_image_url and keep all in normalized_source_images.
• When required URL fields are missing but session-uploaded images are available, auto-fill from normalized uploaded URLs and continue workflow.
• For task_type=product-image, if explicit product_image_url is not a public http(s) URL, force upload normalization first and block generation until URL backfill succeeds.
• For task_type=product-image, source normalization origin must be traceable as one of user_public_url|uploaded_from_session|uploaded_from_local|uploaded_from_markdown|uploaded_from_non_public_url.
• If auto_upload_session_images=true and upload fails, stop the dependent generation stage and return structured missing/failed inputs with retry suggestions.
• For generation tasks, use "preferred fixed tool first; auto-discovery fallback if unavailable"; product-image follows stricter fixed-tool-only rules in subskills/product-image/SKILL.md.
• For logo design, default to aiflow_nougat_create; for product image, default to image_praline_edit_v2.
• For product-image generation, always choose image_praline_edit_v2 as the first and default call target; do not switch to other image-edit tools when image_praline_edit_v2 is available.
• For task_type=product-image or stage=generate-product-image, routing must enter subskills/product-image/SKILL.md as the only orchestration entry.
• For task_type=product-image or stage=generate-product-image, do not bypass subskills/product-image/SKILL.md to call product-image tools directly.
• For task_type=product-image or stage=generate-product-image, first execute mandatory user inquiry and confirmation before any product-image tool call.
• Product-image detailed execution rules (source confirmation, non-URL upload normalization, prechecks, confirmation card, error mapping) are defined only in subskills/product-image/SKILL.md.
• For product-image generation, source image must be user-provided or explicitly user-confirmed in current context; otherwise stop before any tool call.
• If source confirmation precheck fails or image_list does not use the confirmed source URL, return structured_error.code=PRODUCT_IMAGE_UNCONFIRMED_SOURCE and stop before tool call.
• Logo tasks must invoke an actual generation tool and return image links, not text-only plans.
• For 3D generation, default to 3d_hy_image_generate; for single-image input, prioritize fal_tripo_image_to_3d.
• In the discover stage, call GET /api/gateway/mcp/tools first and build arguments from returned tool_schema.
• Gateway tool list may appear under data.list or tools; for product-image task, resolve tool identity by tool_name|skill_name|name and require exact match image_praline_edit_v2.
• If both image_praline_edit_v2 and similarly named tools appear, always select image_praline_edit_v2.
• In the generate stage, call POST /api/gateway/mcp/call with tool_name and args.
• In the upload stage, invoke subskills/upload/SKILL.md and return stable accessible URLs for all local/session images.
• Session-uploaded local images must follow the same upload stage constraints (format/size/accessibility validation and failed-upload retry guidance).
• All Gateway requests must include X-OpenClaw-Key in headers, and keys must be read only from environment variables.
• Before any Gateway call, precheck OPENCLOW_KEY. If missing, return structured_error with code=GATEWAY_KEY_MISSING, stage=precheck, and actionable retry suggestions.
• The report stage must output Markdown consistently and explicitly mark missing fields.
• For product-image tasks, never use local/offline image synthesis fallback (for example PIL scripts or local compositing) to replace Gateway + subskill execution.
• For product-image generation responses, include execution evidence fields: tool_name_used, source_image_url_used, source_image_origin, precheck_report, and non-sensitive call receipt fields from Gateway response.

Language Policy

• All conversational text must follow response_language when provided; otherwise infer from the latest user message.
• If language inference is uncertain, follow the language used in the latest user instruction in the current turn.
• Do not output Chinese by default; Chinese is allowed only when user language is Chinese.
• Do not mix multiple natural languages in one response, except for immutable technical tokens (API paths, field names, tool names).
• Confirmation prompts, error messages, report sections, and suggestions must all use the same target language.
• If any upstream subskill returns text in a different language, rewrite it into the target language before returning.

Failure Handling

• Intra-stage failure: Retry by configuration, up to RETRY_MAX.
• Inter-stage degradation: If one stage fails, continue producing available outputs and document failure reasons in the report.
• Traceability: Record failed stage, input summary, and retry conclusion in the report.
• Product-image precheck failures must return unified structured_error with code, stage, message, details, retry_suggestions.
• Product-image precheck code must use one of PRODUCT_IMAGE_TOOL_UNAVAILABLE|PRODUCT_IMAGE_SOURCE_NOT_ACCESSIBLE|PRODUCT_IMAGE_MISSING_ARGS|PRODUCT_IMAGE_UPLOAD_REQUIRED|PRODUCT_IMAGE_UPLOAD_FAILED|PRODUCT_IMAGE_UNCONFIRMED_SOURCE|PRODUCT_IMAGE_CREATE_TOOL_FORBIDDEN.
• Permission interception: If response is 403 with code=403001, return data.redirect_url directly and prompt the user to activate membership.
• Tool unavailable: If the fixed tool is missing, automatically enable discovery fallback and record the fallback path, except product-image which must follow subskills/product-image/SKILL.md fixed-tool policy.

Security and Compliance

• Never expose keys, tokens, or cookies in outputs.
• Validate image format and size before upload (max 20MB).
• Return only necessary results and publicly accessible links.
• Run the release security validation script before packaging to detect plaintext keys and block packaging.