Jayden-X-L/forkprobe
663Last commit Aug 14, 2026
forkprobe DSH plugin
ForkProbe helps you compare AI skills before committing to one. It submits your task to the model directly (baseline) and several candidate skills, runs them in parallel, and generates a local HTML report showing outputs, costs, and AI evaluations. You choose the winner, and the agent continues along that path.
How to install the forkprobe DSH plugin
dsh plugin --profile web add "github:Jayden-X-L/forkprobeCopying does not run this command. Review the repository and version before installing the forkprobe DSH plugin.
forkprobe DSH plugin data source
forkprobe DSH plugin snapshot date: Aug 16, 2026
discovered
What the forkprobe DSH plugin can do
- Parallel execution of up to multiple candidate skills with a single command
- Generates a local HTML report with outputs, token estimates, file previews, and AI judge comments
- Supports artifact comparisons for presentations, research figures, reports, web pages, and videos
- Provides native DSH tools `forkprobe_compare` and `forkprobe_resume` for seamless integration
- Recommends candidate skills from curated lists, local installations, EverMind Hub, GitHub, or BYO paths
Where the forkprobe DSH plugin fits
- Deciding which writing skill to use for academic polishing, style rewriting, or reviewer response
- Choosing the best skill for generating a PPTX, research figure, research report, or web page
- Comparing video generation skills for product launch videos, motion graphics, or talking-head edits
- Evaluating multiple skills for a research task before committing to one pipeline
- Quickly selecting the right skill for a new task without relying on skill descriptions alone
Who the forkprobe DSH plugin is for
- AI skill developers who want to test their skills against baselines
- Researchers and writers who need to compare multiple AI writing or generation skills
- DSH users who want an integrated skill selection workflow within their agent
forkprobe DSH plugin limitations
- Not suitable for simple deterministic tasks where the correct tool is already known
- Requires DeepSeek Harness (DSH) environment for the native plugin mode
- Currently at version v0.10, still in early development
- External GitHub candidates should be checked for license, dependencies, and output paths before execution
forkprobe DSH plugin: from the repository README
Quoted from the Jayden-X-L/forkprobe README, the upstream source of the forkprobe DSH plugin. Copyright remains with the original authors.
· <a href="./README.en.md">English README</a> · <a href="https://jayden-x-l.github.io/forkprobe/downloads/forkprobe-skill.zip">下载 skill zip</a> · <a href="#deepseek-harness-原生插件">安装 DSH 插件</a> </p> <p align="center"> <img alt="MIT License" src="https://img.shields.io/badge/license-MIT-111827"> <img alt="Version v0.10" src="https://img.shields.io/badge/version-v0.10-2563eb"> <img alt="Local first reports" src="https://img.shields.io/badge/report-local--first-0f9f8f"> <img alt="Agent skill selector" src="https://img.shields.io/badge/agent-skill%20selector-2563eb"> <a href="https://github.com/deepseek-ai/deepseek-harness"><img alt="DeepSeek Harness supported" src="https://img.shields.io/badge/harness-DeepSeek-0f9f8f"></a> <a href="https://github.com/topics/dsh-plugin"><img alt="DSH plugin community" src="https://img.shields.io/badge/community-dsh--plugin-2563eb"></a> <a href="https://github.com/openai/codex"><img alt="Built with OpenAI Codex" src="https://img.shields.io/badge/built%20with-OpenAI%20Codex-111827"></a> </p> ForkProbe 是一个 AI Skill 选型与试跑工具。它会把同一个任务交给模型本身和多个候选 skill，并排试跑，生成本地 HTML report，让你看到真实输出之后再选择 winner。 **v0.10 新增 DeepSeek Harness 原生插件：** DS
Read the full READMERepository license: MIT
forkprobe DSH plugin questions
How do I install ForkProbe as a DSH plugin?
Run `dsh plugin --profile web add "github:Jayden-X-L/forkprobe"` to install it into your DSH web profile. If you also use the headless profile, run the same command with `--profile headless`. After restarting the profile, you can use the `forkprobe_compare` and `forkprobe_resume` tools.
How do I use ForkProbe to compare skills?
After installation, you can tell your DSH agent something like "请使用 ForkProbe 先推荐几个适合这次改写的 Skill，等我确认后再用原生 DSH subagent 并行试跑，打开 Report 让我选择 Winner，并沿胜出结果继续。" Or you can directly call the `forkprobe_compare` tool with `confirmed=true` and a list of skills.
What kind of outputs can ForkProbe compare?
ForkProbe supports comparing text outputs (e.g., writing, polishing), PPTX files, research figures (PNG, SVG, PDF, TIFF), research reports, web pages (HTML with screenshots), and video artifacts (MP4 with previews). Each comparison generates a local HTML report with file previews and AI evaluations.
Can I use ForkProbe with my own custom skills?
Yes, you can add custom skills by providing a path to a local directory, a GitHub repository, or a URL. ForkProbe will scan and include them as candidates. However, you should check the license, dependencies, and output paths of external skills before running them.
How do I choose the winner and continue the task?
The local HTML report includes a "Continue" button. After you review the results and select a winner, clicking that button saves a local handoff and instructs the agent to continue the task using the winning skill. The report also allows you to anonymously share your selection to help the community.