slash-agent brings an LLM coding assistant directly into your terminal—no background processes, no workflow interruption, running locally or via cloud APIs.

What happened: slash-agent is a lightweight shell integration that lets you type /agent in Bash to invoke an LLM (via Ollama, OpenAI, or Azure OpenAI) to diagnose errors, edit files, and run commands. It captures your recent terminal history or tmux scrollback automatically, executes proposed commands in an interactive pseudo-terminal where you can confirm, edit, or refuse each action, and syncs directory changes and environment variables back to your parent shell when done.

Why it matters: Most AI coding tools run in isolated subshells and cannot modify your current working directory or environment. slash-agent solves this by using a state synchronization protocol so the agent's changes (cd, export) carry back into your active session—keeping you in your terminal without breaking workflow. It consumes zero CPU or memory when not in use (no daemons or background processes), and supports both private local models (Ollama) and cloud services.

What to watch: Installation is a single curl command; the tool supports Bash, Zsh, Ksh, and Fish, plus WSL2 on Windows. You can preview agent actions with --dry-run, auto-confirm safe tasks with --yes, or steer each command interactively by typing y/n/e/c at a confirmation prompt. Critical commands (rm -rf, sudo) are not auto-confirmed by default to prevent accidental damage.