/list
Browse and search available inference recipes.
> /plugin marketplace add spark-arena/sparkrun > /plugin install sparkrun@sparkrun
How it fires
How this command gets triggered: by you, by Claude, or both.
- Fires itselfClaude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/list
Context preview
What this command does when you run it.
Browse and search available inference recipes.
Command definition
list.md/sparkrun:list
Browse and search available inference recipes.
Usage
/sparkrun:list [query]
Examples
/sparkrun:list
/sparkrun:list qwen3
/sparkrun:list llama-cpp
Behavior
When this command is invoked:
1. **Browse all recipes** (no filter):
sparkrun list
2. **Browse with filters:**
sparkrun list --all # include hidden registry recipes
sparkrun list --registry <name> # filter by registry
sparkrun list --runtime vllm # filter by runtime (vllm, sglang, llama-cpp)
sparkrun list <query> # filter by name (contains-match)
sparkrun list @<registry> # shorthand for --registry <registry>
sparkrun list @<registry>/<query> # that registry, filtered by <query>
3. **Search for recipes** by name, model, runtime, or description (contains-match):
sparkrun recipe search <query>
sparkrun recipe search <query> --registry <name> --runtime sglang
sparkrun recipe search @<registry>/<query> # registry-scoped shorthand
Use `sparkrun recipe search` as the first attempt when the user wants to find a particular recipe. Only fall back to other approaches if it doesn't return useful results.
4. **Inspect a specific recipe** (by exact name or recipe file):
sparkrun recipe show <recipe>
sparkrun recipe show <recipe> --tp <N> # include VRAM estimate for N nodes
Use `sparkrun recipe show` when given a specific recipe name or file path -- these may not appear in search results.
5. If the user wants to validate or check VRAM for a recipe:
sparkrun recipe validate <recipe>
sparkrun recipe vram <recipe> --tp <N> --max-model-len 32768 --gpu-mem 0.9
6. To export a normalized recipe:
sparkrun recipe export <recipe> # print to stdout
sparkrun recipe export <recipe> --json # JSON format
sparkrun recipe export <recipe> --save out.yaml # save to file
Notes
- Recipes come from built-in and custom registries
- Run `sparkrun registry update` or `sparkrun update` to fetch the latest recipes
- Use `sparkrun registry list` to see configured registries
- `sparkrun list` shows everything; `sparkrun recipe search` filters by query; `sparkrun recipe show` inspects a known recipe
- Recipe names support `@registry/name` syntax for explicit registry selection
Read more
/sparkrun:list
Browse and search available inference recipes.
Usage
/sparkrun:list [query]
Examples
/sparkrun:list /sparkrun:list qwen3 /sparkrun:list llama-cpp
Behavior
When this command is invoked:
1. **Browse all recipes** (no filter):
sparkrun list
2. **Browse with filters:**
sparkrun list --all # include hidden registry recipes sparkrun list --registry <name> # filter by registry sparkrun list --runtime vllm # filter by runtime (vllm, sglang, llama-cpp) sparkrun list <query> # filter by name (contains-match) sparkrun list @<registry> # shorthand for --registry <registry> sparkrun list @<registry>/<query> # that registry, filtered by <query>
3. **Search for recipes** by name, model, runtime, or description (contains-match):
sparkrun recipe search <query> sparkrun recipe search <query> --registry <name> --runtime sglang sparkrun recipe search @<registry>/<query> # registry-scoped shorthand
Use `sparkrun recipe search` as the first attempt when the user wants to find a particular recipe. Only fall back to other approaches if it doesn't return useful results.
4. **Inspect a specific recipe** (by exact name or recipe file):
sparkrun recipe show <recipe> sparkrun recipe show <recipe> --tp <N> # include VRAM estimate for N nodes
Use `sparkrun recipe show` when given a specific recipe name or file path -- these may not appear in search results.
5. If the user wants to validate or check VRAM for a recipe:
sparkrun recipe validate <recipe> sparkrun recipe vram <recipe> --tp <N> --max-model-len 32768 --gpu-mem 0.9
6. To export a normalized recipe:
sparkrun recipe export <recipe> # print to stdout sparkrun recipe export <recipe> --json # JSON format sparkrun recipe export <recipe> --save out.yaml # save to file
Notes
- Recipes come from built-in and custom registries
- Run `sparkrun registry update` or `sparkrun update` to fetch the latest recipes
- Use `sparkrun registry list` to see configured registries
- `sparkrun list` shows everything; `sparkrun recipe search` filters by query; `sparkrun recipe show` inspects a known recipe
- Recipe names support `@registry/name` syntax for explicit registry selection
sparkrun - launch, manage, and stop LLM inference workloads on NVIDIA DGX Spark systems. Live support: https://discord.com/invite/GH5kRgv6ZD

