Catalog PyPI
PACKAGE EXPLORERView in registry

agent-worktree-manager

PyPI

Manage isolated Python environments and Git worktrees across projects

Versions collected1
Direct dependencies8From release 0.2.0
Last observed21 Oct, 08:57 UTC

README

Version 0.2.0 documentation

Agent Worktree Manager

Working on several branches or with coding agents means keeping track of both source checkouts and their Python environments. awm pairs Git worktrees with isolated environments and provides a terminal UI to browse and clean them up across projects.

Start with an environment that already works. AWM reproduces its installed dependencies and points selected editable packages at new worktrees. Other editable packages keep their original source links. Install AWM once, separately from your project environments.

Supports Python 3.11–3.14 on macOS and Linux; Git is required. Standard venv/pip is the default; uv and conda backends are optional. AWM and your project environments can use different Python versions.

Installation

From PyPI, once the first release is published, choose either:

uv tool install agent-worktree-manager
# Or install into your chosen Python environment:
python -m pip install agent-worktree-manager

To install from a source checkout now, run at this repository's root:

uv tool install --editable .
# Or:
python -m pip install -e .

To activate a sandbox in your current terminal from the TUI, enable shell integration:

eval "$(awm shell-init zsh)"  # use bash for Bash

Add the same line to ~/.zshrc (or ~/.bashrc) to enable it in new terminals. Then press s on a sandbox to close AWM and activate its environment.

Register your first project

From your project's Git checkout, register its existing Python environment:

awm init --name my-project --base /absolute/path/to/base-env --repo app=.
awm doctor
awm create my-task --repo app
awm run my-task -- python --version
awm

--base is the environment directory, not its Python executable. Use --backend uv for a uv base or --backend conda for a conda base. Repeat --repo ALIAS=PATH to include more repositories. Registration writes awm.toml and machine-specific settings in git-ignored .awm/, and adds the project to your global overview. From elsewhere, select it with awm --project my-project.

Use with coding agents

The bundled awm skill guides agents through creating, resuming, inspecting, and cleaning up worktrees with their environments. Install it for either agent:

# Codex
awm skill --install "$HOME/.agents/skills/awm"
# Claude Code
awm skill --install "$HOME/.claude/skills/awm"

In Codex, select it through /skills or mention $awm. In Claude Code, use /awm. For example: “create a sandbox for fixing the login tests and work there.” See the agent guide for custom locations and updating a skill.

Contributing

Please follow Conventional Commits for commit messages, for example feat(tui): add sandbox shell shortcut or fix(env): preserve editable paths. These messages drive automated versioning and release notes.

License

BSD 3-Clause — Copyright (c) 2026, Jan Wyzula.

Documentation

Usage · Configuration · AI agent skill

Downloads & popularity

Awaiting observations

No metric observations yet

Download counts and popularity will appear here as the collector receives them.

Versions 1

Collected releases
VersionPublishedDirect dependenciesOpen version
0.2.010 Sept 20268

Package advisories

0
No matching advisories collected

This describes our collected data; it is not a security assessment.

Metadata history

Meaningful changes
Current collected state Current

Observed since 21 Oct, 08:57 UTC

No superseded metadata states collected. Unchanged observations do not create new revisions.

Record identity

package:580bdcf31f47614be001d6afd0c425bd7bf6ef87f9610488b9c50f309bcdca18