agentic-dev-toolkit

Tooling for doing agentic development across as many machines as I work on: the laptop in front of me, and any box I reach over SSH and want to keep working after I disconnect. One repo, one install script, kept in sync everywhere with git pull; everything is installed by symlink, so nothing ever gets copied and left to drift out of date. The idea started after watching this video of the german YouTuber Niklas Steenfatt.

vibe. The CLI at the center of it: one git branch, one worktree, and one tmux session per task. It hands a task off between machines through git alone — start a task on one box, park it, attach from another — and vibe loop runs a task unattended, checking in on a schedule instead of babysitting a terminal. vibe doctor diagnoses the setup when something looks wrong.

Agent Skills. A growing set of portable SKILL.md skills, written to the open Agent Skills format rather than tied to one agent — things like keeping a repo's HANDOFF.md and PROJECT_STATUS.md current across sessions, running a non-behavioral codebase health check, or staging an unattended loop that drives a pull request to green and mergeable.

Shared config across agents. A single global memory file — the local/server split, the handoff discipline, where each kind of information belongs — gets installed into Claude Code, Codex, and Gemini's own instruction paths from one source file, so the same working agreement holds no matter which agent a given machine is running.