DSH Marketplace

forkprobe

Jayden-X-L/forkprobe

Compare multiple skills on the same task and pick the winner.

755PythonMITSource

Install

Add forkprobe to DeepSeek Harness

Install into

via GitHub · GitHub source

Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.

What happened when we ran it

Installed cleanly when we ran it

Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 11d agoon DeepSeek Harness 0.2.2-rc.1.

Show it in your README

install verified — dshmarketplace

For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.

Due diligence

Before you install forkprobe

  • Source of record: Jayden-X-L/forkprobe — present in the community registry that DSH's own plugin market installs from.
  • Licensed under MIT.

Security

What installing forkprobe runs

Build scripts

No build script was blocked in our run.

Source: our sandbox install on DeepSeek Harness 0.2.2-rc.1, 11d ago.

Install hooks

package.json declares no preinstall, install or postinstall script.

Source: package.json on GitHub, read at our last sync. prepare is not recorded; it runs only on a GitHub install, and shows up under build scripts when pnpm blocks it.

README mentions

shell commands (bash, exec, spawn …)

Source: a keyword match on the README — a hint about what the plugin touches, not an analysis of its code.

Not an audit. These are the record of one install and two text scans; nobody has read this plugin's code on your behalf. Plugins run with your agent's permissions.

The AI take

What it is — It runs multiple skills in parallel on the same task, generates a local HTML report, and lets you choose the winner to continue execution.

Who it is for — In academic polishing, SCI writing, PPTX generation or image prompt design tasks, having the agent compare multiple skill outputs in parallel can help you pick the winner. For tasks where the answer or tool path is already clear, people who can execute directly can skip it because direct execution will be faster.

Watch out — Sandbox testing passed; it was installed in a fresh profile and registered into it. Static checks indicate it will execute shell commands, but no other issues found. Installation from source requires manually allowing the build script.

The verdict — I would install it because it turns skill selection into an observable parallel execution process, but only when deliverables need file previews and QA.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.

What forkprobe does

Parallel task paths are compared and converge into one selected result with a local report.

forkprobe is a DeepSeek Harness plugin that compares several skills on the same task and helps select a winner. It runs a baseline and multiple candidate skills with the same input, then opens a local HTML report containing the outputs, elapsed time, estimated token use, file previews and AI review suggestions. In DSH, the native forkprobe_compare tool starts candidate subagents in parallel. After selection, the report’s “Continue” action creates a local handoff and returns the chosen result so the current agent can continue with the winning skill.

Candidates can come from the local installed-skill set, EverMind Skill Hub, GitHub or a BYO path. The repository also documents direct runs with python3 scripts/compare.py, using options such as --input, repeated --skill, --judge and --output. The plugin scans local skills, removes duplicates and can optionally support anonymous winner sharing.

This is intended for tasks where output quality or artifact quality is uncertain, including writing, PPTX, figures, research reports, web pages and video. It is a poor fit for simple, deterministic work where the correct tool path is already clear. External GitHub candidates still require checks for licence, dependencies and output paths. The terminal surface reaches the environment to run comparison scripts and candidate pipelines, which may create reports and generated files with the agent’s permissions; the README does not describe an additional isolation layer. DSH users do not need a nested dsh process, and the plugin does not copy DSH credentials.

forkprobe documentation

How it behaves

forkprobe treats skill selection as a comparison run. It recommends a small candidate set, runs baseline and the selected skills with identical input, and writes a local HTML report. Depending on the task, the report can include text, previews, source files, QA results, media metadata, estimated token use, elapsed time and AI review comments. The documented workflow ends with a user-selected winner and a continuation handoff.

In DSH, the native forkprobe_compare tool starts candidate subagents in parallel. The README states that this native implementation does not launch a second dsh process or copy DSH credentials. It also retains installation confirmation, local Skill scanning, candidate de-duplication and optional anonymous winner sharing.

Commands

The README shows a direct comparison command:

python3 scripts/compare.py --input /tmp/forkprobe-input.txt --skill baseline --skill writing-anti-ai --skill humanizer-zh --skill remove-ai-flavor-writing-skill --judge --output /tmp/forkprobe-report.html

--input takes an input file, --skill can be supplied for each candidate, --judge enables the documented AI review step, and --output specifies the report path. The DSH-native tool is named forkprobe_compare; its full argument schema is not included in the supplied README.

Candidate sources and outputs

Candidates may be recommended from the curated directory, locally installed Skills, EverMind Skill Hub, GitHub or a BYO path. The documented report types include text outputs, PPTX files, PNG/SVG/PDF/TIFF figure assets, research evidence files, runnable web pages, screenshots, MP4 media, subtitles, transcripts and source code, depending on the comparison mode.

Known limits

Image-generation comparison is listed as planned rather than supported. For external GitHub candidates, the README recommends checking licence, dependencies and final output paths before execution. Simple deterministic tasks are explicitly outside the intended use case.

Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub

Same category

Alternatives to forkprobe

7.0k

One repository, three parts: a runtime injector for DSH plugin packages (inject, hot-reload, unload, promote a dev staging tool to the front, route self-heal, plus a settings-page plugin manager that lists, unloads and drags folders in to internalize), a task-aware reasoning-mode router agent preset (router-standard / router-spec / router-react), and a graded two-level task protocol whose six tools (commit_star, lock_stage, revise_do, edit_plan, mark_task, redteam_verdict) pin task state to disk. The injector implementation ships in-tree, so the install carries its own behaviour rather than a dependency list.

Installs, but needs a build approved first

GitHub sourceJavaScript1mo ago

AI review

dsh-routing-suite

What it is — A DSH suite that ships a runtime injector, task-aware reasoning-mode presets, and a graded two-level protocol whose six tools pin task state to disk.

Who it is for — You are on DSH 0.1 preview with DeepSeek Pro or Flash, running spec-then-execute work that needs the agent to switch reasoning modes by task and to write stage state to disk. Skip it when your DSH is already outside >=0.1.0-rc.6 <0.2.0; the preview still breaks APIs, so the version window is the constraint, not the job type.

Watch out — In our run the install started, then pnpm blocked the build script and DSH never finished registration. Until that build is allowed by hand, the plugin does not actually land.

The verdict — I'll take it only after unlocking the blocked build: the injector ships in-tree, so that allow step is how you accept this runtime's behaviour rather than a dependency list.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
Source

mirage

strukto-ai

3.7k

Swaps the filesystem and bash providers for a mirage virtual workspace: file tools and shell commands run over mounted resources (RAM, S3, Redis, Slack, Gmail, Notion, Postgres) instead of the host disk, with per-mount read/write/exec modes, per-command sandbox routing (monty, pyodide, quickjs in process; docker, e2b, daytona remote), and installed CLIs (git, gh, slack, linear, ntn, gws, or one you register) as head words in the virtual terminal.

No one-line install — this plugin lives inside a larger repository and publishes no npm package.

GitHub sourceTypeScript24d ago

AI review

mirage

What it is — mirage provides a unified virtual filesystem for AI agents, allowing bash tools to operate across mounted resources.

Who it is for — It is suitable for AI agents that use bash tools and need to access multiple backends such as S3 and Slack at the same time. It is not suitable for agents that only run on local filesystems without remote mounted services.

Watch out — No obvious issues found. There is a static check prompt indicating it will execute shell commands. Sandbox testing was not performed.

The verdict — I would not install it because sandbox testing was not performed.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Details

oh-my-agent

first-fluke

1.4k

Runs project OMA checks before and after tool execution and when a turn tries to finish. Installs its OMA CLI dependency and initializes unconfigured projects on first use. Requires Node.js 22+; initial setup needs network access and project write access.

No one-line install — this plugin lives inside a larger repository and publishes no npm package.

GitHub sourceTypeScript24d ago

AI review

oh-my-agent

What it is — oh-my-agent runs project OMA checks before and after tool execution and when a turn tries to finish, installing its OMA CLI dependency and initializing unconfigured projects on first use.

Who it is for — It fits users who need to run project checks before and after tool execution and when a turn tries to finish, as well as project initialization on first use. Users who avoid network access and project write permissions during setup can choose not to install it.

Watch out — Initial setup requires network access and project write access. It executes shell commands. Sandbox testing has not been performed, no obvious pitfalls found.

The verdict — I won't install it because it requires network access and project write access during initial setup and has no sandbox testing.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Source
386

Manages local DeepSeek Harness skills in Settings and shows shared Agent skills as read-only.

Installed cleanly when we ran it

npm packageJavaScript19d ago

AI review

dsh-skills-manager

What it is — Manages local DeepSeek Harness skills in Settings and displays shared Agent skills as read-only.

Who it is for — Users who need to manage local and project skills across multiple tools in DeepSeek Harness setups would use this plugin. Users who only rely on shared agent skills without local or project management do not need it.

Watch out — Passed sandbox test: installed in a new profile and registered by Harness. No obvious issues found. Compatible with DeepSeek Harness 0.2.x series.

The verdict — I would install it because it supports incremental updates from public GitHub repositories with backup restoration.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
Details

oh-dsh

hust-open-atom-club

323

Community distribution: TUI, desktop, and Web UI as one bundle with layered installation.

GitHub sourceTypeScript2mo ago

pi2dsh

weijiafu14

212

Pi Host ABI compatibility engine: after one install, unmodified Pi extensions from npm mount as native DSH plugins with `dsh plugin add <pi-package>`. Verified end to end on stock DSH with pi-mcp-adapter (full MCP manager: OAuth, resources, prompts, MCP Apps, elicitation, sampling), @tintinweb/pi-subagents, pi-code, pi-hermes-memory and pi-background-tasks; `pi2dsh inspect` reports a package's compatibility before installing.

DeepSeek Harness 0.2.2-rc.1 refuses the latest release

npm packageTypeScript21d ago

AI review

pi2dsh

What it is — pi2dsh is a Pi host ABI compatibility engine that mounts unmodified Pi npm packages as native DSH plugins on DeepSeek Harness.

Who it is for — Pi plugin developers who want to run their npm packages unmodified on DeepSeek Harness should install it. Developers who are not involved in Pi plugin development do not need to install it.

Watch out — It was tested successfully in a new profile and registered into the profile. A restart of DeepSeek Harness is required. Installation may require manually approving build scripts if pnpm blocks them.

The verdict — I would install it if I need Pi ecosystem plugins on DeepSeek Harness because it provides unmodified compatibility.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Source