Show and tell: visionDS —— DeepSeek Harness 多提供商视觉/OCR 技能(可安装组合包)/ a multi-provider vision/OCR skill for DeepSeek Harness #384
deveuper
started this conversation in
Show and tell
Replies: 1 comment
|
这个思路很好 - 把已有 skill 转成 dsh 可安装 bundle,正是官方文档缺的「如何打包发布插件」路径的实操。我们手册第 7 章(生态)记录了这类社区资产,已把 visionDS 收录(含 dsh 插件开发第 4 章的可克隆模板,纯函数 + waterfall + 测试的组织方式): https://github.com/Electricitysheep/dsh-handbook/blob/main/docs/07-ecosystem.md 如果后续想把这个 bundle 的打包步骤写成教程,欢迎来手册共建。 |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
visionDS: a multi-provider vision/OCR skill for DeepSeek Harness
I converted a ZCode vision skill into a DeepSeek Harness skill and published it as an installable bundle, following the Package and install a plugin tutorial.
Repository: https://github.com/deveuper/visionDS (public, MIT, no API keys)
Install
It is plain JavaScript with no
preparestep, so the git install needs no build allowance. The bundle's plugin reads its bundledSKILL.mdand registers thevision-dsskill at apply time viactx.skills.registerwith a directoryresourceBase, so the skill's Python scripts resolve inside the installed package.Manual install also works for any SKILL.md-compatible agent: copy the
skill/directory to~/.agents/skills/vision-ds.What it does
When the main model has no vision (e.g. DeepSeek text-only), the skill sends images/screenshots/URLs to a configured vision provider and returns the text result:
--provider windows-ocr/macos-ocr.%APPDATA%\vision-ds\.env,~/.config/vision-ds/.env, orVISION_DS_CONFIG_DIR); the repository ships no secrets.Requires Python 3.10+ for the recognition script.
Feedback welcome — happy to adjust anything that does not match Harness conventions.
All reactions