Skip to content

Registry commands ​

Search, install, and publish packages from the kdeps registry, and install, update and remove kdeps assets (harness prompts, tool definitions, commands, themes, templates, schemas, model registries). An argument naming an asset set (tools) or item (tools/bash_exec) is an asset; anything else is a package.

Commands ​

Each is a top-level kdeps command; --registry <url> (or KDEPS_REGISTRY_URL) points it at another registry.

CommandDescription
kdeps searchSearch for packages in the kdeps registry
kdeps infoShow what a kdeps asset is for and where it runs (kdeps info tools/read_file, kdeps info events), or the metadata and README of a package, component, agent, or GitHub repo
kdeps installInstall assets (set, set/name[@version], with what they require), or a package from the registry, a GitHub repo (owner/repo; github:owner/repo when the owner is named like an asset set), or a local archive (.kdeps .kagency .komponent)
kdeps uninstallRemove assets, or an agent or component installed from the registry
kdeps updateUpdate assets (no args: the profile and every installed asset; --check lists updates), or an installed agent or component
kdeps listList installed and local components and agents, each with the tokens its agent-mode tool definition adds to the LLM context; --assets lists every asset with its version, state, shelf and context tokens; kdeps list <shelf> shows one shelf
kdeps submitGenerate a registry formula YAML for submitting a package via GitHub PR
kdeps verifyRun LLM-agnostic verification on a package directory
kdeps profileChoose what assets kdeps installs: minimum, workflows, agent, deploy, full, custom, or sets and items; no name opens the picker. Items the new profile leaves out are removed

LLM context tokens ​

kdeps install, kdeps update and kdeps list estimate what each item adds to the model's context (about 4 bytes per token). A package installed from the registry is a tool in agent mode, so its name, description and input schema count:

text
$ kdeps install scraper-native
✓ Installed scraper-native (workflow) @1.0.0
  Web scraping + AI summarization using the native scraper executor.
  LLM context: about 96 tokens as an agent-mode tool (its tool definition)

Examples ​

bash
kdeps search scraper
kdeps install scraper-native
kdeps install scraper-native@1.0.0
kdeps install ./my-component-1.0.0.komponent
kdeps list
kdeps info scraper-native
kdeps info tools/read_file           # an asset: its purpose, where kdeps runs it, its state
kdeps uninstall scraper-native
kdeps update scraper-native
kdeps submit --tag v1.2.0
kdeps verify .

# kdeps assets
kdeps install tools/bash_exec      # one asset, with what it requires; prints ~tokens it adds to the LLM context
kdeps update                       # every installed asset to its newest version
kdeps update --check               # what would change
kdeps uninstall themes/vim
kdeps list --assets                # version, state, shelf, ~context tokens
kdeps list rag                     # one shelf; the id stays tools/embedding_search
kdeps profile agent                # install the agent profile

Publishing a package ​

The kdeps registry is GitHub-hosted. Packages live in the author's own GitHub repo; the registry indexes a formula file per package.

bash
# 1. Tag a release in your repo
git tag v1.2.0 && git push --tags

# 2. Generate the formula YAML (downloads the tarball and computes SHA256)
kdeps submit --tag v1.2.0

# 3. Open a PR to https://github.com/kdeps/registry
#    adding the printed formula as formulas/<your-package-name>.yaml

Formula file format ​

yaml
# formulas/my-agent.yaml
name: my-agent
version: 1.2.0
type: workflow
github: owner/my-agent-repo
tarball: https://github.com/owner/my-agent-repo/archive/refs/tags/v1.2.0.tar.gz
sha256: <computed-by-kdeps-registry-submit>
description: ...
tags: [llm, chat]
license: Apache-2.0

kdeps install downloads from the GitHub tarball URL and verifies the SHA256 locally.

See also ​

Released under the Apache 2.0 License.