Introduction

There are a few ingredients left in the fridge, and I don’t know what to cook. This is a problem many people face daily. Current solutions usually involve opening a multimodal model dialog, manually uploading images, describing requirements, and organizing prompts yourself, repeating this process every time.

BigFood encapsulates this workflow into a skill following the Agent Skills open standard. Once installed in DeepSeek Harness (dsh), “Snap fridge → Identify ingredients → Recommend recipes” becomes an ability native to the agent, eliminating the need to rebuild prompts every time. Below is an introduction to the skill’s positioning, functionality, and installation method.

What is This

BigFood (package name @kobenfang/dsh-bigfood, version 1.0.0), maintained by kobenfang, is positioned as an “AI Fridge Butler”: upload ingredient images to identify ingredients and recommend recipes based on the results. It also supports multi-image mixing, fridge ingredient management, and purchase reminders.

It follows the Agent Skills open standard (SKILL.md + YAML frontmatter), so it is not tied to dsh alone. Frameworks like OpenClaw and Claude Code, which support Agent Skills, can also use it. The license is MIT.

Core Features

Verified features are as follows:

  1. Upload ingredient images to identify them, provided the model used supports multimodal (image analysis);
  2. Recommend recipes based on identified ingredients;
  3. Multi-image mixing, meaning multiple images can be uploaded simultaneously for combined analysis;
  4. Fridge ingredient management (inventory management), recording existing ingredients at home;
  5. Purchase reminders, combining inventory status to prompt what needs to be bought.

Full feature descriptions are written in SKILL.md within the repository, and the repository directory structure lists only this core file:

bigfood/
├── SKILL.md          # Skill definition (Agent Skills standard)

Installation and Activation

The official installation method is to clone the repository to the Agent Skills directory; there is no dedicated command like dsh plugin add. It is recommended to put it in the user-level shared directory so it can be used by all projects:

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

If you only want to use it in a specific project, you can put it in the project’s .agents directory:

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

Choose either method. After completion, start dsh, and you will find bigfood in the skill list.

OpenClaw users can also install from ClawHub:

npx clawhub install bigfood

If you want to confirm this package on ClawHub, you can execute:

npx clawhub search bigfood

Typical Usage

After installing and starting dsh, the entire workflow is: upload ingredient images, the skill calls multimodal capabilities to identify the ingredients, and then gives recipe recommendations based on the results. If ingredients are scattered in several places at home, multiple images can be uploaded at once for mixed analysis; combined with fridge ingredient management and purchase reminders, you can also get prompts when stock is insufficient.

Search keywords listed in the repository README include: 冰箱 (fridge), 菜谱 (recipe), 食材 (ingredients), 做饭 (cooking), 吃什么 (what to eat), recipe, ingredient, cooking, which can serve as a reference when locating this skill in the skill directory. For more details, view SKILL.md directly in the repository.

Applicable Scenarios and Notes

Suitable for two types of people: those who are dsh users wanting to add a practical “image-to-cook” skill to their agent; and Agent Skills ecosystem developers who can refer to it for organizing an image input skill using SKILL.md.

There are two points to note before using:

  1. Ingredient recognition depends on multimodal model support. If the current dsh session is connected to a pure text model, the image recognition step will not work.
  2. The skill will run with the permissions of the current dsh process. It is recommended to read the repository source code and SKILL.md before installing to ensure the behavior meets expectations; the license is MIT, with relatively loose terms, but it is still recommended to confirm for yourself.

Conclusion

BigFood turns the high-frequency, small need of “what is in the fridge and what can be made” into an out-of-the-box skill. Installation is just a single git clone, and it is not affected by changing frameworks—this is precisely the value of the Agent Skills standard.

Directory Page: https://www.skillhub.cn/plugins/kobenfang/BigFood (Community-maintained site, no official affiliation with DeepSeek / Hypothesis)

GitHub Repository: https://github.com/kobenfang/BigFood