Docker via WSL (Windows)

SkillCloud & infra

Use when YOU (the AI agent) are running on Windows OUTSIDE WSL (Git Bash/MSYS/PowerShell shell) and need to run ANY docker / docker compose command. Docker Desktop runs on the WSL2 engine, so commands must be re-issued INSIDE WSL via wsl.exe -- running them from the Windows shell on a network/SMB drive (Z:, UNC) corrupts bind-mount paths. Does NOT apply if your shell is already inside WSL. Triggers on: docker, docker compose, docker-compose, container, bind mount, volume, 'is a directory', mount source wrong, Windows + Docker Desktop, WSL.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Docker via WSL (Windows) skill

What this skill tells your AI

The instructions your AI receives, as published by netresearch/docker-development-skill in skills/docker-via-wsl/SKILL.md and read by ahel’s review.

When this applies

This applies when you, the AI agent, are running on a Windows host and your shell is OUTSIDE WSL -- your Bash tool is Git Bash/MSYS (uname -s shows MINGW64…/MSYS…) or you are in PowerShell/cmd. The docker binary there talks to Docker Desktop, but the daemon and its filesystem live in WSL2, so you must run commands inside WSL via wsl.exe.

If your shell is already inside WSL (uname -s shows Linux, native path /home/<user>/...), this skill does not apply -- run docker directly.

The problem

On Windows, Docker Desktop's daemon runs inside the WSL2 VM. Issue every docker command -- run, build, exec, pull, push, volume, network, inspect, compose, … -- from inside WSL, against a native Linux path. Never from a Windows shell (Git Bash/MSYS/PowerShell), especially on a mapped network/SMB drive (Z:, UNC \\host\share).

Why it matters

Driving Docker from a Windows shell whose CWD is on a network share forces Docker Desktop to translate the Windows bind path (e.g. Z:\proj\config) into a VM path. On SMB/UNC drives this is unreliable: it can duplicate a path segment (e.g. …/user/user/proj/config), so Docker mounts a wrong, empty directory and auto-creates the missing bind source as an empty directory. A file the container expects (init.sql, a config file) then appears as a directory -> could not read from input file: Is a directory. Your editor writes still land on the correct path via the share, so host and container views silently diverge.

This is NOT a Docker cache bug nor a "single-file bind mounts are fragile" problem (single-file binds work fine from WSL). The root cause is piloting Docker from Windows/SMB instead of from WSL.

The rule

Run any Docker command through WSL, in a native Linux path:

wsl.exe -e bash -lc "cd /home/<user>/<project> && docker compose up -d"
wsl.exe -e bash -lc "docker ps"
wsl.exe -e bash -lc "docker build -t myimg /home/<user>/<project>"

Keep the project on the Linux filesystem the WSL distro sees natively (/home/<user>/...), not browsed through the Windows drive letter.

Diagnosis and fix

See references/diagnosis-and-fix.md for inspecting the mounted path, comparing host vs container views, and cleaning up a phantom bind directory.

Signals

GitHub stars
20
Forks
1
Last commit
Sep 2026

ahel review

  • S4info
    community integration — published by netresearch, not docker

Automated review, not a security audit. Ruleset v1.

Advanced
Catalog kind
skill
Gateway key
docker-via-wsl
Source
github.com/netresearch/docker-development-skill