dekk 1.3.0

Reusable environment detection, activation, and wrapper tooling for developer CLIs.

dekk

Make project CLIs runnable with one config and zero manual activation.

dekk turns project setup into a reusable runtime layer. Declare the environment once in .dekk.toml, then use dekk to detect tools, activate the environment, install wrappers, and make project commands work on fresh machines without hand-written setup steps.

Install And Start

Recommended for end users:

pipx install dekk

Fallback if you already manage Python packages directly:

python -m pip install --upgrade dekk

After installation:

• Use dekk for the CLI.
• Use python -m dekk as a fallback if your scripts directory is not on PATH yet.

The Problem

Every project needs environment setup: conda environments, PATH entries, environment variables, tool versions. Developers manually activate things. AI agents waste thousands of tokens describing setup steps. CI pipelines duplicate configuration.

The Solution

Declare your environment once in .dekk.toml. dekk handles detection, activation, wrapper generation, and installed command setup from that single source of truth.

[project]
name = "myapp"

[conda]
name = "myapp"
file = "environment.yaml"

[tools]
python = { command = "python", version = ">=3.10" }
cmake  = { command = "cmake", version = ">=3.20" }
cargo  = { command = "cargo" }

[env]
MLIR_DIR = "{conda}/lib/cmake/mlir"

[paths]
bin = ["{project}/bin"]

Three Pillars

1. Detect

Zero-dependency detection of your entire development environment:

• Platform: OS, architecture, Linux distro, WSL, containers
• Package managers: conda/mamba, with environment validation
• Build systems: 25+ (Cargo, CMake, npm, Poetry, Maven, Gradle, ...)
• Compilers: GCC, Clang, Rust, Go with versions and targets
• CI providers: 14 (GitHub Actions, GitLab CI, Jenkins, ...)
• Shells: 9 types (bash, zsh, fish, tcsh, PowerShell, ...)
• Workspaces: Monorepo detection with dependency graphs

from dekk import PlatformDetector, CondaDetector, BuildSystemDetector

platform = PlatformDetector().detect()
# PlatformInfo(os='Linux', arch='x86_64', distro='ubuntu', ...)

conda = CondaDetector().find_environment("myenv")
# CondaEnvironment(name='myenv', prefix=Path('/opt/conda/envs/myenv'))

builds = BuildSystemDetector().detect(Path("."))
# [BuildSystemInfo(system=BuildSystem.CARGO, root=Path("."), ...)]

2. Activate

Read .dekk.toml, resolve conda paths, set environment variables, validate tools:

from dekk import EnvironmentActivator

activator = EnvironmentActivator.from_cwd()
result = activator.activate()
# ActivationResult(env_vars={'CONDA_PREFIX': '...', 'MLIR_DIR': '...'}, ...)

Or from the CLI:

$ eval "$(dekk activate --shell bash)"

On Windows PowerShell:

PS> Invoke-Expression (& dekk activate --shell powershell | Out-String)

3. Wrap

Generate a self-contained launcher that bakes in the full environment. This is what makes dekk zero-friction.

$ dekk wrap myapp ./bin/myapp
  Generated myapp -> /path/to/project/.install/myapp

$ myapp doctor    # just works -- no activation needed

On POSIX, the wrapper is a simple shell script with hardcoded paths:

#!/bin/sh
export CONDA_PREFIX="/home/user/miniforge3/envs/myapp"
export PATH="/home/user/miniforge3/envs/myapp/bin:$PATH"
export MLIR_DIR="/home/user/miniforge3/envs/myapp/lib/cmake/mlir"
exec "/home/user/miniforge3/envs/myapp/bin/python3" \
     "/home/user/projects/myapp/tools/cli.py" "$@"

On Windows, dekk installs a .cmd launcher in the project's .install directory so the command works from both Command Prompt and PowerShell without requiring Activate.ps1.

From Python:

from dekk import WrapperGenerator

result = WrapperGenerator.install_from_spec(
    spec_file=Path(".dekk.toml"),
    target=Path("tools/cli.py"),
    python=Path("/opt/conda/envs/myapp/bin/python3"),
    name="myapp",
)

Installation

pipx install dekk
python -m pip install --upgrade dekk
python -m pip install --upgrade "dekk[tracking]"
python -m pip install --upgrade "dekk[all]"

First Run

The default path should be simple:

dekk --help
dekk doctor
dekk init --example quickstart

That gives you a working CLI immediately, a system check, and a starter .dekk.toml in the current directory.

If dekk is not found yet, use python -m dekk --help immediately.

If you want a built-in starter without writing files yet:

dekk example quickstart
dekk example conda --output .dekk.toml

Built-in templates live in examples/.dekk.toml.quickstart, examples/.dekk.toml.minimal, and examples/.dekk.toml.conda.

Typical next steps:

# Python CLI from a repo with pyproject.toml
dekk install ./tools/cli.py

# POSIX shells
eval "$(dekk activate --shell bash)"

# Install a launcher after your project builds a target
dekk install ./bin/myapp --name myapp

# Remove a launcher again
dekk uninstall myapp

For Python scripts, dekk install ./tools/cli.py uses pyproject.toml to create or refresh .venv automatically on first run. For binaries and conda-backed projects, dekk install uses .dekk.toml to bake the required environment into the installed command. When the install directory is not already on PATH, dekk also appends the project's .install directory to the active shell config when it can, then asks for a shell restart once.

Use dekk uninstall <name> to remove an installed launcher. Pass --remove-path when you also want dekk to remove that project's PATH export from the shell config.

# PowerShell
Invoke-Expression (& dekk activate --shell powershell | Out-String)
dekk install .\dist\myapp.exe --name myapp

Naming Conventions

dekk uses one name per surface area:

• PyPI package: dekk
• Python import: dekk
• CLI command: dekk
• Project config file: .dekk.toml
• Default wrapper location: .install/ in the current project

That keeps installation, imports, command usage, and project setup distinct and predictable.

CLI Framework

dekk includes a production-quality CLI framework built on Rich and Typer. Use it as the foundation for your own CLI tools:

from dekk import Typer, Option

app = Typer(
    name="myapp",
    auto_activate=True,      # auto-setup from .dekk.toml
    add_doctor_command=True,  # built-in health check
    add_version_command=True, # built-in version info
)

@app.command()
def build(release: bool = Option(True, "--release/--debug")):
    """Build the project."""
    ...

if __name__ == "__main__":
    app()

Styled output

from dekk import print_success, print_error, print_warning, print_info
from dekk import print_header, print_step, print_table

print_header("Building MyApp")
print_step("Compiling...")
print_success("Build complete!")
print_warning("Debug symbols not stripped")

Progress indicators

from dekk import spinner, progress_bar

with spinner("Installing dependencies..."):
    install_deps()

with progress_bar("Processing", total=100) as bar:
    for item in items:
        process(item)
        bar.advance()

Structured errors

from dekk import NotFoundError, DependencyError

raise NotFoundError(
    "Compiler not found",
    hint="Install the required toolchain for this project",
)
# Displays styled error with hint, exits with code 3

Multi-format output

from dekk import OutputFormatter, OutputFormat

fmt = OutputFormatter(format=OutputFormat.JSON)
fmt.print_result({"status": "ok", "version": "1.0"})

LLM-friendly subprocess runner

from dekk import run_logged

result = run_logged(
    ["cargo", "build", "--release"],
    log_path=Path(".logs/build.log"),
    spinner_text="Building...",
)
# Shows spinner, captures output to log, prints path for agents to read

.dekk.toml Reference

[project] -- required

[project]
name = "myapp"
description = "Optional description"

[conda] -- conda/mamba environment

[conda]
name = "myapp"
file = "environment.yaml"

[tools] -- required CLI tools

[tools]
python = { command = "python", version = ">=3.10" }
cmake  = { command = "cmake", version = ">=3.20" }
ninja  = { command = "ninja" }
cargo  = { command = "cargo", optional = true }

[env] -- environment variables

[env]
MLIR_DIR = "{conda}/lib/cmake/mlir"
LLVM_DIR = "{conda}/lib/cmake/llvm"
MY_HOME  = "{project}"

Placeholders: {project} (project root), {conda} (conda prefix), {home} (user home)

[paths] -- PATH prepends

[paths]
bin = ["{project}/bin", "{project}/target/release"]

Examples by Language

Python + Conda

[project]
name = "ml-pipeline"

[conda]
name = "ml-pipeline"
file = "environment.yaml"

[tools]
python = { command = "python", version = ">=3.10" }
jupyter = { command = "jupyter" }

[env]
PYTHONPATH = "{project}/src"

Rust

[project]
name = "my-rust-app"

[tools]
cargo = { command = "cargo", version = ">=1.70" }
rustc = { command = "rustc" }

[paths]
bin = ["{project}/target/release"]

C++ with CMake

[project]
name = "physics-sim"

[conda]
name = "physics-sim"
file = "environment.yaml"

[tools]
cmake = { command = "cmake", version = ">=3.20" }
ninja = { command = "ninja" }
clang = { command = "clang", version = ">=17" }

[env]
CMAKE_PREFIX_PATH = "{conda}"

Node.js

[project]
name = "web-app"

[tools]
node = { command = "node", version = ">=18" }
npm  = { command = "npm" }

[env]
NODE_ENV = "development"

Go

[project]
name = "api-server"

[tools]
go = { command = "go", version = ">=1.21" }

[env]
GOPATH = "{home}/go"

[paths]
bin = ["{home}/go/bin"]

For AI Agents

dekk reduces environment setup from 2000-5000 tokens to ~150 tokens:

Before (what agents had to explain):

Check if conda is installed. If not, install miniforge. Create environment with conda env create -f environment.yaml. Activate with conda activate myenv. Set MLIR_DIR to the conda prefix. Export LD_LIBRARY_PATH...

After:

Run myapp install. The wrapper handles everything.

Detection API Summary

Module	What it detects
PlatformDetector	OS, arch, distro, WSL, containers, package manager
CondaDetector	Conda/mamba environments, packages, validation
BuildSystemDetector	25+ build systems with targets and workspaces
CompilerDetector	GCC, Clang, Rust, Go with versions and targets
CIDetector	14 CI providers with git metadata and runner info
ShellDetector	9 shell types with config files and capabilities
WorkspaceDetector	Monorepos with dependency graphs and build order
DependencyChecker	CLI tool versions against constraints
VersionManagerDetector	pyenv, nvm, asdf, rbenv, rustup
LockfileParser	7 lockfile formats across ecosystems

Architecture

dekk is organized in three tiers:

• Tier 1 (Core): Foundational detection and config modules. Platform, conda, deps, workspace, config, remediation.
• Tier 2 (Extended): Paths, build systems, compilers, shells, toolchains, versions, CI.
• Tier 3 (Frameworks): Diagnostics, commands, scaffolding.

The CLI framework ships in the base dekk install.

Documentation

• Getting Started
• Architecture
• .dekk.toml Specification
• Wrapper Generation
• Examples by Language
• Contributing

License

MIT