Back to Home
Uncategorized August 16, 2026

New Tool: dsh-plugin-dev-skills

If you spend time building extensions for the DeepSeek Harness (DSH) ecosystem, you know how repetitive the setup can feel—creating manifests, wiring event handlers, aligning with agent APIs, and packaging for distribution. A new agent skill called dsh-plugin-dev-skills aims to shrink that friction by providing a ready‑made scaffold for plugin development that works across several […]

If you spend time building extensions for the DeepSeek Harness (DSH) ecosystem, you know how repetitive the setup can feel—creating manifests, wiring event handlers, aligning with agent APIs, and packaging for distribution. A new agent skill called dsh-plugin-dev-skills aims to shrink that friction by providing a ready‑made scaffold for plugin development that works across several coding assistants.

What You Need to Know

The skill is installed as an agent capability that you invoke from your preferred coding companion—Claude Code, Codex, the native DSH interface, VS Code Copilot, or any agent that follows the same skill protocol. Once activated, it offers a set of commands that generate the boilerplate files required for a DSH plugin: a manifest.json, an entry‑point script, stubs for event listeners, and a basic test harness. The generated structure follows the official DSH plugin specification, so you can start coding the unique logic of your extension right away.

Beyond file generation, the skill includes helpers for common tasks such as validating the manifest against the schema, running a local test server that mimics the DSH host, and creating a distributable package (a zip or npm‑style bundle) with a single command. Because the skill is agent‑agnostic, you can switch between coding assistants without relearning the workflow; the same set of commands works in each environment.

Why It Matters

Developers often lose time to repetitive setup steps that do not add value to the core functionality of a plugin. By automating those steps, dsh-plugin-dev-skills lets you focus on the domain‑specific code—whether that is implementing a new LLM adapter, handling custom events, or integrating with external services. The reduction in boilerplate also lowers the chance of configuration errors that can cause plugins to fail during host validation.

Because the skill works with multiple agents, teams that use different tools can maintain a shared plugin development process. This consistency simplifies onboarding for new contributors and makes it easier to maintain a single source of truth for plugin templates across projects.

Key Details

  • Command set: init, validate, test, package, and publish.
  • Supported agents: Claude Code, Codex, DSH native, VS Code Copilot, and any agent implementing the Agent Skills interface.
  • Output format: Generates a manifest.json, main.js, events.js, and a basic Jest test file.
  • Local test server: Starts a mock DSH host on port 8080 for rapid iteration.
  • Packaging: Produces a zip file ready for upload to the DSH marketplace or internal repository.

What’s Next

The skill is open for community contributions; the roadmap includes adding templates for specific adapter types (such as vector store connectors) and integrating with CI pipelines to automate validation on each push. Users are encouraged to file issues or submit pull requests on the project’s repository to shape future features.

📌 Source: Github

Related Articles

Uncategorized August 19, 2026

Proactive Road Safety Intervention in Australia: Predicting Risky Driving Hotspots from Connected Vehicle Data

Transport agencies in Australia have long depended on crash reports to spot dangerous roads, a method that only reveals problems

Uncategorized August 19, 2026

A decodability criterion predicts when hidden-state selection beats majority voting in large language models

When a language model generates several answers to the same prompt, the usual way to pick a final response is

Uncategorized August 19, 2026

DiSCO: Defending text-to-image generation through distribution-guided contrastive prompt optimization

Recent advances in text‑to‑image models have unlocked impressive creative capabilities, but they also open the door to unsafe outputs such