Foreword

DSH’s philosophy is “everything is a plugin,” and capabilities are extended through skills. However, the skill ecosystem mostly consists of developer-oriented tools. If one day you ask an agent, “The pressure is low today, what bait should I use for crucian carp?”, a generic model response is often too broad because it lacks structured fishing knowledge.

The Agent Skills Open Standard offers a simple solution: package domain knowledge into a SKILL.md and a few data files, which are automatically loaded when the framework starts, eliminating the need to paste materials into the prompt every time. The BigFish introduced below is a fishing skill implemented according to this standard.

What is this

BigFish (kobenfang/bigfish) is a fishing assistant skill for DeepSeek Harness (DSH), created by kobenfang, with package name @kobenfang/dsh-bigfish, version 1.0.0, and License MIT (both README and package.json indicate this).

It solves specific problems: fishing spot analysis, fish condition analysis, and sharing fishing spots. It can analyze fish conditions by taking photos of the water surface and, combined with weather and pressure, recommend fish species, fishing methods, and bait.

It follows the Agent Skills Open Standard (SKILL.md + YAML frontmatter), so besides dsh, it can also be used on frameworks like OpenClaw and Claude Code that support this standard.

Core Features

Verified capabilities include:

  • Fishing spot analysis
  • Fish condition analysis: Analyze fish conditions by taking a photo of the water surface
  • Fishing spot sharing
  • Recommend fish species, fishing methods, and bait based on weather and pressure
  • Supports Lure fishing, Feeder fishing, Wild fishing, and Sea fishing
  • Automatically record catches and fishing trip reports
  • Purely knowledge-base driven, no script dependencies

The last point is worth expanding on. The entire skill does not rely on any scripts; the knowledge is all stored in 4 YAML files. The directory structure is as follows:

bigfish/
├── SKILL.md          # Skill definition (Agent Skills standard)
├── scripts/knowledge/bait-recipes.yaml
├── scripts/knowledge/fish-species.yaml
├── scripts/knowledge/lure-guide.yaml
├── scripts/knowledge/spot-analysis.yaml

These correspond to bait recipes, fish species, lure guides, and spot analysis respectively. Complete functional descriptions are in SKILL.md in the repository, and the README explicitly states “see SKILL.md for details,” so you can read it directly after installation.

Installation and Enablement

The README does not provide an installation command in the form of dsh plugin add. The official way is to clone the repository into the skills directory. There are two methods:

Method 1, placing it in the user-level shared directory (Recommended):

mkdir -p ~/.agents
git clone https://github.com/kobenfang/bigfish.git ~/.agents/bigfish

First create the ~/.agents directory, then clone the repository into it. This way, all projects can share this skill.

Method 2, placing it in the project directory:

mkdir -p .agents
cp -r bigfish .agents/

Suitable for situations where you only want to enable it in a specific project.

After installation, bigfish can be found in the skill list when dsh starts.

If you are using OpenClaw, you can also install it from ClawHub:

npx clawhub install bigfish

For searching:

npx clawhub search bigfish

Typical Usage

  1. After installation, start dsh and confirm that bigfish is loaded in the skill list.
  2. Since recommending fish species, methods, and bait requires combining weather and pressure, the README suggests using weather-class tools. That is, configure the weather-class tools in dsh first, and then use this skill for the recommendations to have real-time data to base on.
  3. After that, you can use it according to the skill’s positioning: take a photo of the water surface for it to analyze fish conditions, ask about suitable fish species, methods, and bait for the current weather and pressure, or ask it to record catches and generate trip reports. Specific interaction details are subject to SKILL.md in the repository.

Applicable Scenarios & Notes

Who is it suitable for:

  • Fishing enthusiasts who use dsh daily, covering Lure, Feeder, Wild, and Sea fishing styles;
  • Developers who want to reference how a “pure knowledge-base skill” is written—writing no scripts and using YAML to organize domain knowledge is an example that can be read directly from the source code.

A few notes:

  1. The skill runs with the permissions of the current dsh process. Before installing any third-party skill, it is recommended to check the source code and license first. BigFish is MIT, and the repository only has 5 files, so the cost to read it once is very low.
  2. Weather and pressure-related recommendations rely on external weather data. This capability is limited if used alone; configure the weather-class tools as mentioned above before using it.
  3. The repository address in the README installation command is lowercase kobenfang/bigfish. GitHub addresses are case-insensitive, pointing to the same repository as the page address kobenfang/BigFish.

Conclusion

The value of BigFish does not lie in technical complexity—it is just 4 YAMLs plus a SKILL.md—but in demonstrating a lightweight extension method: structuring a vertical domain’s knowledge and handing it to an agent. For fishing players, it gives dsh an on-the-go advisor; for skill authors, it is a ready-to-copy knowledge-base skill template.

Repository address: https://github.com/kobenfang/BigFish

Community directory page: https://www.skillhub.cn/plugins/kobenfang/BigFish (This directory is an independent community site with no official affiliation to DeepSeek / Huafang)