Agent plugins for DeepSeek Harness
Skills, personas and sub-agents for specific jobs.
Agent plugins shape how the agent behaves on a particular kind of work. A skill packages instructions for one task, a persona sets tone and priorities, and a sub-agent takes a piece of the job and handles it separately. Most of what these plugins ship is prompt text, sometimes with a few tools attached.
3 plugins
- nuwa-skillalchaincyf Give Nuwa a name and it researches, extracts and blind-tests a person skill. It follows the open Agent Skills standard, so it is not limited to DSH. Agents 38.1k
- awesome-gpt-image-2freestylefly A library of image generation prompts, templates, and reusable skills for agent workflows. Its gallery supports prompt discovery, while a separate 2.5 section documents shared-prompt comparisons. Agents 35.8k
- Distillytitanwings Distilly converts materials about a person into a reusable Person Profile, packaged as an Agent Skill. It supports colleague, relationship, and public or fictional figure workflows. Agents 26.5k
How to choose
Because the substance is the prompt, read it. A good one is specific about when it applies and what it should produce; a vague one will change the agent's behaviour in ways you cannot predict. Check which model it was written and tested against, since instructions tuned for one model can work poorly on another. For sub-agents, look at how many extra model calls a run makes, because that is where cost and waiting time come from.
Terms used on this page
- Skill
- A reusable set of instructions, sometimes with scripts or files, that the agent loads for a specific task.
- Persona
- A system prompt that fixes the agent's role, tone and priorities for a session.
- Sub-agent
- A separate agent run, with its own context, that the main agent hands part of a task to.
Questions and answers
What is the difference between a skill and a persona?
A skill is loaded when a matching task comes up and covers how to do that task. A persona applies to the whole session and covers how the agent should act in general. They can be used together.
Do sub-agents use more tokens?
Yes. Each sub-agent runs its own conversation with the model, so a task split three ways costs roughly three sets of requests. The trade is a cleaner main context and, sometimes, parallel work.