weshopai/weshop-dsh-plugin/inspect-weshop-canvas

Inspect weshop Canvas

Non-negotiable insertion rule

• Add in the canvas is exclusively a human-facing material importer. Never use browser automation to insert a generated output through it.
• Every successful image MUST be inserted with weshop_canvas_publish_result. Publish generated video, audio, or text with weshop_canvas_publish_asset and the matching mediaType before replying that the task is complete.
• Publishing the result is part of generation completion, not an optional follow-up step. Do not ask the user to add it manually.
• If publishing fails, report the failure and retry the canvas tool; do not fall back to the webpage Add button.

Visible execution progress

Use weshop_canvas_report_progress at meaningful transitions so the canvas exposes concise, verifiable status — not private chain-of-thought. Set one stable startedAt timestamp for the task and include a monotonic percent:

1. interpreting (~8%): one-sentence understanding plus outputPlan.
2. researching (~12–20%, only when required): report the platform/market/source focus.
3. planning (~22%): selected WeShop commercial workflow or model and why it fits.
4. prompt-ready (~32%): set promptStatus to "已完成"; summarize the prompt constraints.
5. generating (35–85%): report the active model and expected output count.
6. publishing (~92%): report that completed files are being added automatically.
7. complete (100%) or error: report the concrete outcome.

Do not emit updates for every API poll. Keep progress useful and sparse.

• Call weshop_canvas_get_state before reasoning about the full canvas.
• Call weshop_canvas_get_selection when the user says "this", "these", "selected", or otherwise refers to the active selection. Always inspect selectedItemIds and selectedItems; they contain the complete multi-selection. The singular fields are compatibility aliases for the primary/last-selected item only.
• Treat kind: material as source/reference input and kind: result as generated output.
• Use each item's provenance to explain how it entered the canvas. Do not infer missing provenance.
• If connected is false, ask the user to open WeShop for DeepSeek Harness first.
• Call weshop_canvas_get_requests when the user asks to handle an operation submitted from the canvas right-click menu.

Canvas context-menu requests

• reverse-prompt: inspect the request image, infer a detailed reusable prompt, report it in the conversation, then call weshop_canvas_complete_request with the prompt as the summary.
• upscale: run the most suitable image enhancement/upscale capability (see the weshop-openapi skill), publish the output with weshop_canvas_publish_result, then complete the request.
• local-edit: use the request's edit instruction and source image; preserve the source item, publish the edited output as a new result, then complete the request.
• Never complete a request until its analysis or generation has actually finished.

WeShop OpenAPI workflow

When the user asks to generate, edit, transform, or animate canvas content:

1. Read the canvas state and current selection first.
2. Treat all selected material items as candidate API inputs, preserving their order from selectedItems. For agents supporting multiple reference images, pass the complete compatible selection. Do not use unrelated results unless the user requests them.
3. Load the weshop-openapi skill and follow its workflow. It owns intent compilation, endpoint selection, uploads, request parameters, polling, and API-key handling.
4. If the WeShop API key is unavailable, stop before any API request. Tell the user to open the WeShop canvas and click Configure API Key in its top bar, then retry; alternatively they may set WESHOP_API_KEY before starting Harness. Direct them to https://www.weshop.ai/apiKey only to obtain a key.
5. Preserve output provenance for canvas insertion: WeShop agent name/version, execution ID, source canvas item IDs, task name or prompt, creation time, and returned asset URL.
6. Classify generated outputs as result; never overwrite or reclassify their source material items.
7. After a successful generation, publish the matching asset type directly from the returned URL so it appears automatically: images use weshop_canvas_publish_result; video/audio/text use weshop_canvas_publish_asset. For every output of the same execution, pass batchId equal to the execution ID and a zero-based batchIndex in returned-output order so the canvas keeps the batch together. Do not download remote results first; use a local path only for a genuinely local-only output. Never use Add for this.
8. Put complete creation lineage in provenance: method, WeShop agent and version, executionId, prompt or taskName, sourceItemIds, returned URL, and completion timestamp.

Pure plugin boundary

• The weshop canvas is a canvas-only surface inside the DeepSeek Harness GUI and does not embed a separate chat.
• Read canvas state through the native tools, then converse with the user in the existing DeepSeek conversation.
• Never expose the WeShop API key to the browser; all WeShop requests run inside the Harness host process, and the key is supplied through the Harness environment.

The native Cordis plugin is the authority for canvas context. The weshop-openapi skill is the authority for WeShop generation behavior.