jayden-x-l/forkprobe

Don’t guess which AI skill is useful—just look at the results side by side.

Release page · English README · Download the skill zip · Install the DSH plugin

ForkProbe is an AI skill selection and trial run tool. It assigns the same task to the model itself and multiple candidate skills, running them side by side, generating a local HTML report, so you can see the real output before selecting the winner.

v1.1 Added image prompt / style direction comparison: ForkProbe now allows comparison of image prompts / style pipelines. Each candidate is prompt.md, [T1]], [T2]], [T3]], and render-notes.md, without calling the image API inside the runner; When Codex is running and the host has image generation capability, optional rendering validation can be performed based on local render-queue.json, while other agents can use external user rendering to backfill rendered.png.

After selecting a winner, the “Continue” button in the report simultaneously saves the local handoff, allowing the Agent to continue the task based on the winning skill. Users can choose whether to anonymously share their current Skill selection in the same area, recommending a priori accumulation samples for future communities.

As the number of skills on the network increases, the question is no longer “do you have a skill?” but “which skill should you use for the current task?” ForkProbe’s goal is straightforward: first lay out the results, then let the Agent continue working along the path you selected.

When should you use ForkProbe?

• You’re unsure which skill to use for the current task and want to see the actual damage before deciding.
• You want to compare the baseline with multiple skills, rather than just trusting the description of the skill.
• Your deliverables are PPTX, research figure packages, research reports, image prompt/style packages, and can run web pages or video videos, requiring document review, preview, and Q&A.
• You want to find candidates from the locally installed Skill, EverMind Skill Hub, GitHub, or BYO paths, and run a small test run again.
• Not suitable for simple deterministic tasks: If the answer or toolpath is already clear, direct execution is faster.

How it works

ForkProbe turns skill selection into an observable process:

1. Recommend a small number of candidate skills or artifact pipelines from the curated directory, locally installed skills, EverMind Skill Hub, GitHub, and BYO paths.
2. Use the same input to run the baseline and multiple candidates.