This article is derived from the plugin's already-indexed fields (wiki / faq / compatibility_json / readme), not freshly generated by AI. Source field is noted at the end of each section.

Quick start

dsh-image-bridge

— source: plugin_wiki.wiki_content

Install & verify

dsh plugin --profile web add @kbpoyo/dsh-image-bridge

Run the command above in your DSH Web Profile. Then enable the plugin in the plugin list.

— source: plugins.install

Key points

• 一键桥接：纯文本模型下，输入框粘贴图片后出现「桥接发送图片」按钮，点击即可发送，模型能读到图片。
• 回车兜底：不点按钮、直接回车发送图片，也会在发送入口自动改写为桥接标记发送，与按钮等效。
• 多模态直通：多模态模型下按钮自动隐藏、图片原生发送，插件零干预。
• 即时感知：会话中切换模型，按钮立即显隐，不闪烁、不跳动。
• 自动清理：桥接产生的临时图片定时自动清理，无需手动管理。

— source: plugin_wiki.readme_en (fallback readme_raw)

FAQ

Do I need to configure anything after installation?

No, you don't need to. The config in cordis.patch.yml defaults to empty; after npm installation, simply restart dsh web to take effect; in text-only model sessions, pasting an image will show a "Bridge Send Image" button (cordis.patch.yml:1-5 / README.md:25-27).

What image formats and sizes can be bridged?

Follows attachments.imageLimits: supports PNG / JPEG / WebP / GIF (other types fall back to .png), single and total size limits, and max images per message are determined by the host config; the button performs full validation against limits at onClick (lib/index.js:35-43 / lib/index.js:214-236).

Where are the temporary images stored? Will they stay on disk permanently?

Stored in a hidden directory under the current session's workspace <cwd>/.dsh-image-bridge/ (starts with a dot, not shown in regular file managers); the plugin cleans up files older than 24 hours or when the directory exceeds 512 MB (oldest first), and also cleans up on plugin uninstall (lib/index.js:27-28 / lib/index.js:76-100 / README.md:76-79).

Will the button disappear after switching to a multimodal model?

Yes. The client subscribes to the official model directory store (event-driven) and adds a 1-second fallback poll; the button hides immediately when switching; for multimodal models, images are sent directly via native, the plugin doesn't intervene (lib/client.js:131-151 / README.md:12).

Can the bridged model really "see" the images? Do I need to manually select a vision tool?

It depends on whether the model itself has vision/OCR/grounding tools in its directory—the plugin only delivers the image and writes in the text marker "Check current available skills and tools first before calling". You don't need to specify a tool; if the model directory has no vision capability, the model will honestly tell you so (lib/index.js:151-159 / README.md:73).

What if the button keeps showing for a custom provider's text-only model?

The adapter only reports as multimodal when its model declaration contains input: ['text','image']; custom models without input declared are always reported as text-only, the plugin faithfully reads this. To hide the button, you need to add the input field in the model declaration or use adapter-managed directory (README.md:79-89).

Can I switch back to a text-only model after sending an image?

No, this is a DSH native limitation (session logs with native images cannot switch back), you need /compact or start a new session; this plugin doesn't bypass this limitation (README.md:81-82).

Does the bridging override parent-to-child agent image sending?

No, it doesn't override. Messages with images sent from parent agent to child agent don't go through session.prompt, they still go through the native gate MODEL_DOES_NOT_SUPPORT_IMAGES (README.md:91 / lib/index.js:300-305).

— source: plugin_wiki.faq_json

Compatibility

• DSH: >=0.1.0-rc.6
• Node: 未声明

— source: plugin_wiki.compatibility_json

Pitfalls

Review the upstream repo before installing. This guide is auto-derived from indexed fields and may lag the latest release. If anything contradicts the official docs, treat the upstream source as authoritative.

— source: general rule