Adds a path-backed attachment dock for images, documents, archives, code, and other files, leaving file reading to the Agent's available tools.
Install
# from npm (prebuilt)
dsh plugin --profile web add dsh-paste-to-path
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:Johnny-xuan/dsh-paste-to-path
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time. Only install sources you trust, and pin a commit (github:owner/repo#sha).
README
A universal attachment dock for DSH.
English | 简体中文
dsh-paste-to-path adds a general-purpose attachment Dock to the DSH Web composer.
Paste or drop images, PDFs, Word and Excel documents, archives, code, logs, and other files, then review and manage them together before sending.
The DSH 0.1.0-rc.6 Web composer natively accepts PNG, JPEG, WebP, and GIF. PDFs, Office documents, archives, and other formats do not currently have the same unified attachment entry point. Even an image may fail when the selected model does not support image input or the active adapter is text-only.
dsh-paste-to-path does not extend the model's native content types. It takes a simpler route:
File
↓
DSH Host
↓
Local path
↓
Agent
↓
Your own tools
Send images to your own vision tool, PDFs to a document reader, and archives to shell or extraction tools.
The plugin owns attachment intake, management, and path delivery. Your Agent tool stack decides how to read the file.
Path flow at a glance
What it does
Universal attachment Dock
Paste or drop a file and an attachment card appears above the composer.
Each card shows the file name, size, category, and path. You can remove it before sending.
Images also have thumbnails and lightbox previews.
One flow for many file types
Images, PDFs, Office documents, code, logs, archives, and other binary files all use the same attachment flow:
File → Save on Host → Path reference → Agent
You do not need a separate model-input protocol for every file type.
Turn long pasted text into an attachment
Normal text still pastes normally.
When pasted content crosses the configured threshold, the plugin can save it as a .txt file instead of putting tens of thousands of characters directly into the composer.
The default threshold is 8000 characters.
Edit text files before sending
Text and code attachments can be edited directly in the Dock when they are below the configured size limit.
The default limit is 1 MiB.
Open files with the system application
For a local DSH deployment, host.openPath can open an attachment with the system's default application.
Installation
Install into a DSH Web profile:
dsh plugin --profile web add dsh-paste-to-path
Before the npm release, you can also install directly from GitHub:
dsh plugin --profile web add github:Johnny-xuan/dsh-paste-to-path
Restart DSH after installation:
dsh web
The package includes a dsh.bundle manifest, so the required loader patch is loaded with the plugin.
How it works
When you paste or drop a file into the composer, the plugin catches it and saves it on the DSH Host:
<workspace>/.dsh/pastes/<category>/
The composer does not contain the file's contents. It keeps an attachment reference instead.
When you send the message, DSH's reference codec expands that reference into a short path instruction:
Paste / drop file
│
▼
Save on DSH Host
│
▼
Attachment Dock
shows the file card
│
▼
Send message
│
▼
reference codec
creates a path instruction
│
▼
Agent receives the path
│
▼
Uses an available tool to read it
The implementation uses DSH's extension mechanisms:
conversation.input.dock- input-trigger reference codec
No DSH core modification is required.
What the Agent receives
The plugin does not put file bytes in the initial model request.
For example, a PDF expands to:
Document attachment: /absolute/path/to/report.pdf
Read it using an appropriate tool for this file format.
An image expands similarly:
Image attachment: /absolute/path/to/image.png
Inspect it using an available image-reading method.
Text, code, archives, and other formats receive equivalent path instructions.
The instructions do not require a particular tool. The Agent decides what to do next from the tools that are actually available in the current session.
Bring your own tools
dsh-paste-to-path does not parse file contents.
Connect whichever tools fit your Agent environment, for example:
- images → your own
read_imageor vision tool - PDF / Word / Excel → document readers
- scans → OCR
- code / logs → shell or filesystem tools
- ZIP / TAR → archive extractors
The plugin does not install these tools or assume that the current model has their capabilities.
Any compatible tool that can access the file path on the DSH Host can read the stored attachment.
If the Agent has no suitable tool, the file still enters the Dock and is saved on the Host, but the Agent cannot understand its contents.
Why paths
In DSH's native attachment path, supported file formats and model capabilities are closely related.
In 0.1.0-rc.6, the Web composer currently accepts:
- PNG
- JPEG
- WebP
- GIF
Other MIME types do not enter the same native image path.
An image that passes the format check may still fail if it reaches a model without image input or a text-only adapter.
dsh-paste-to-path separates two concerns:
Give the file to the Agent
and:
Understand the file's contents
The plugin handles only the first.
The file becomes an ordinary file on the Host. The Agent's tool layer handles the second.
Configuration
Default configuration is provided by cordis.patch.yml:
- insert:
- id: paste-to-path
name: dsh-paste-to-path
config:
longTextAsAttachment: true
longTextThreshold: 8000
maxBytes: 26214400
editableTextMaxBytes: 1048576
| Option | Default | Description |
|---|---|---|
longTextAsAttachment |
true |
Save long pasted text as a .txt attachment |
longTextThreshold |
8000 |
Character threshold for long-text conversion |
maxBytes |
25 MiB | Maximum size of one attachment |
editableTextMaxBytes |
1 MiB | Maximum text-file size editable in the Dock |
Override the paste-to-path entry in your profile's cordis.patch.yml to change these values.
File storage
With a workspace, attachments are stored under:
<workspace>/.dsh/pastes/<category>/
Without a workspace, storage falls back to:
$DSH_HOME/tmp-paste/<category>/
File permissions are:
0600
Removing an attachment from the Dock removes only its reference from the current draft. It does not delete the file from disk.
The path therefore remains valid for undo, re-send, or later reference.
Privacy
Files are uploaded from the browser to your own DSH Host and saved on the Host's local filesystem.
The plugin itself does not:
- upload files directly to a model provider
- upload files to a third-party file service
- parse file contents during upload
The initial model request contains only the file path and a short instruction.
If the Agent later uses another tool or external service to process the file, that tool's own behavior and configuration apply.
Design boundary
dsh-paste-to-path owns only this part:
File
↓
Attachment Dock
↓
Host filesystem
↓
Path reference
This part:
Path
↓
Vision / PDF Reader / OCR / Shell / ...
belongs to the Agent's tool layer.
The plugin therefore does not:
- modify or replace a model adapter
- pretend that the model supports vision
- bind the workflow to a fixed vision, OCR, or document tool
- parse attachment contents during transport
- create a native DSH
imagecontent block
Compatibility
Tested with:
DeepSeek Harness 0.1.0-rc.6
DSH is currently a developer preview. Changes to its extension interfaces may require a corresponding plugin update.
Links
More in this category
liustack/modlens★ 1582
Vision bridge for text-only models: paste an image, get structured JSON evidence (OCR, layout, semantics).
superdesigndev/treg★ 412
Tool catalog for agents: search ~2,600 external endpoints (SEO and SERP, backlinks, social, people and company enrichment, ad libraries, scraping) by the task you want done, read each one's parameters and per-call price, then call it with the credential injected server-side. Ships the skill plus an MCP row that stays disabled until TREG_TOKEN is set.
Anionex/dsh-vision-toolkit★ 395
Vision tasks for text-only models: intent-aware image Q&A, long-screenshot OCR, UI reproduction, grounding, and pixel diff.
zhaoolee/notes★ 141
Export DSH conversations as Smartisan Notes-style PNGs, or create and update Markdown notes in a configured account-scoped workspace.
Lum1104/dsh-browser★ 129
Chrome sidebar extension that lets DSH operate your browser directly, no vision capabilities required.
liustack/modsearch★ 101
Web search bridge for text-only agents: ask the web or X, get structured JSON evidence (search, fetch, citations).