A repo that makes my AI workflows reproducible and manageable: skills, commands, MCP servers, prompts, and containerized agent pipelines.
stow/opencode/.config/opencode/ deployed to ~/.config/opencode via Stow
skills/ SKILL.md folders — agent auto-discovers these
commands/ opencode slash-commands — user-invoked, deterministic
opencode-base.jsonc repo-managed config (plugins, permissions)
AGENTS.md global agent guidelines
containers/ every container definition in this repo, one subfolder per
tool/task — dev tooling, MCP servers, and pipeline images alike
<name>/ distrobox.ini or Dockerfile + README (setup, pinned version)
pipelines/ agent task automation that ISN'T a container definition
dagger/ Dagger pipeline code (Go/Python/TS) — references an image
under containers/<name>/ by path, doesn't nest it here
scheduling/ docs on what's scheduled + why (the timer itself lives in
the separate Guix repo as a Guix Home Shepherd service)
docs/ knowledge base: reference docs, architecture diagrams, notes
stow -d stow -t ~ opencode
Symlinks stow/opencode/.config/opencode into ~/.config/opencode.
Then, on every machine that uses this repo, set an environment variable so
OpenCode loads the repo-managed config as a layer without overwriting the
machine's own local ~/.config/opencode/opencode.jsonc:
export OPENCODE_CONFIG=~/.config/opencode/opencode-base.jsonc
Add this to your shell rc or Guix Home config.
dev-browser (browser automation for agents, via Playwright/Chromium):
npm install -g [email protected]
dev-browser install
dev-browser install-skill --agents
Writes its skill file to ~/.agents/skills/dev-browser/