slash-agent

What does this agent do, and when should you use it?

slash-agent is a command-line LLM copilot that lives inside your active Bash, Zsh, Ksh or Fish session: a shell wrapper (bin/slash-agent.sh) registers an /agent command that hands off to a Python orchestrator (slash_agent/main.py). It scrapes the current tmux pane scrollback or shell history as context, then calls a local Ollama instance, OpenAI, Azure OpenAI, or any OpenAI-compatible endpoint such as llama.cpp or vLLM. Proposed commands run through a real pseudo-terminal behind a y/n/e/c confirmation loop, so you can see colored output, progress bars and answer sudo prompts interactively. On exit a temporary sourcing script pushes cd and export changes back into the parent shell. The tool is entirely on-demand with no daemon or idle memory use; native Windows is unsupported and WSL2 is required there.

Running /agent or agent triggers the shell wrapper, which captures context (up to AGENT_TMUX_LINES, default 50 lines of tmux scrollback, falling back to AGENT_HISTORY_COMMANDS, default 20 history entries) and invokes slash_agent/main.py. The orchestrator connects to an LLM using AGENT_BACKEND (ollama, openai, azure_openai, dummy) plus AGENT_ENDPOINT and AGENT_MODEL, with optional AGENT_THINKING_LEVEL, AGENT_TEMPERATURE and AGENT_TOP_P. Each returned command is shown as Confirm action: [y]es / [n]o / [e]dit / [c]omment before being executed in the PTY bridge; -n/--dry-run previews the plan, -y/--yes skips routine confirmations, and --unsafe-yes also auto-confirms critical commands such as rm -rf or sudo. After execution the exit code, PWD and environment updates are written to a sync file that the parent shell sources. It also discovers SKILL.md files from .agent/skills/, .claude/skills/, .github/skills/ and global skill directories.

1. A developer whose build or test suite just failed types /agent Fix this and lets the agent read the traceback in the tmux scrollback and install the missing dependency.
2. Someone setting up a local database, docker compose file or systemd service asks /agent for the commands and approves each one instead of looking up flags.
3. An engineer in an offline or regulated environment points AGENT_BACKEND at local Ollama so terminal logs and source code never leave the machine.
4. A user who wants directory and environment changes to persist in the main shell relies on the cd/export parent-shell sync instead of an isolated subshell.
5. A cautious user previews the agent's plan with /agent -n before letting anything touch the system.

How do you install or deploy this agent?

Install with pipx and register shell integration:

pipx install slash-agent && slash-agent init --install

Or with pip:

pip install slash-agent && slash-agent init --install

Then restart the shell or reload it manually:

source ~/.bashrc

If you prefer not to use pipx, a self-contained curl installer is available:

curl -fsSL https://raw.githubusercontent.com/akatzmann/slash-agent/master/bin/installer.sh | bash

For development from source, clone the repo, install requirements and source the shell script:

pip install -r requirements.txt
source ~/.slash-agent/bin/slash-agent.sh

Fish users source ~/.slash-agent/bin/slash-agent.fish; on macOS use ~/.bash_profile or ~/.zshrc. Windows users must install inside WSL2.

How do you use this agent?

Configure the backend, by default in ~/.config/slash-agent/env (override with SLASH_AGENT_CONFIG_FILE):

export AGENT_BACKEND="ollama"
export AGENT_MODEL="gemma4:latest"
export AGENT_ENDPOINT="http://127.0.0.1:11434"
export OPENAI_API_KEY=""
export AGENT_TMUX_LINES=50
export AGENT_HISTORY_COMMANDS=20
export AGENT_THINKING_LEVEL="off"

Or run the interactive setup wizard, pre-filled with current values:

/agent --configure

Then invoke it anywhere in the terminal:

/agent Fix this crash
/agent -n setup a docker compose file for PostgreSQL and Redis
/agent -y update package lists and install tree

Update to a new release with:

/agent --update

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ