CREAIT-nl

Dsh Plugins — AI skill for Claude Code

AI community

Plugins for DeepSeek Harness: deep research as an agent preset, per-model generation limits, Claude Code hook compatibility, and web fetch/search tools.

How to install Dsh Plugins

This entry records only its repository, not the path inside it, so there is no exact command to give. Open CREAIT-nl/dsh-plugins and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Dsh Plugins does

Plugins for DeepSeek Harness: deep research as an agent preset, per-model generation limits, Claude Code hook compatibility, and web fetch/search tools.

Alternatives in AI

  • EchoBird — One-click install + model switch:Claude Code,Codex CLI (OpenAI), Grok Build (xAI), DeepSeek Harness, Kimi Code 3.1k ★
  • TokenTracker — Local-first AI token usage & cost tracker for 31 coding tools incl 1.4k ★
  • Loushang — AI-native agent harness for coding workflows by python: multi-model LLM orchestration, stateful sessions, tool 1.3k ★

README

dsh-plugins

Plugins for [DeepSeek Harness](https://github.com/deepseek-ai). One repo, nine packages, published independently under `@creait`. They came out of running dsh against self-hosted models, and each closes a gap the harness leaves open.

Package What it does Why
hookkit
@creait/dsh-hookkit
Config-driven lifecycle hooks, including context injection. Declare event → handler → outcome in YAML. dsh has the seams (agent/pre-step, tools/pre-execute, …) but reaching them means writing a plugin. This turns them into config.
gen-limit
@creait/dsh-gen-limit
Per provider/model concurrency cap, enforced at subagent spawn and at the stream, with a settings card. One self-hosted GPU has a real ceiling; a fan-out of subagents will find it. Denying the spawn beats letting the child die opaquely.
web-search-searxng
@creait/dsh-web-search-searxng
Points web_search at a self-hosted SearXNG instance instead of the native route. Search without handing every query to a third party.
web-fetch
@creait/dsh-web-fetch
A guarded local fetch provider behind ctx.web, with SSRF blocking on the resolved address. dsh implements web_fetch in full but ships no provider, so the tool is present and every call fails.
research-mode
@creait/dsh-research-mode
Deep research as an agent mode: a fixed, reviewed loop that plans, researches in adaptive parallel rounds, synthesises and reviews. A loop the model rewrites per call re-earns the same mistakes per call. Shipping it as a preset also keeps it out of every session that is not research.
think-level
@creait/dsh-think-level
A reasoning effort per provider and model, plus one control to change it for the session you are in. dsh's effort dropdown is per session and remembered nowhere, so "this model should