Commands
Every slash command, the modes you can toggle, and how to write your own commands.
Type / in the input box to open the command palette.
Built-in commands
| Command | Description |
|---|---|
/pause |
Block new requests and tool calls; admitted work, including permission waits, may finish |
/unpause |
Allow new requests and tool calls; continue queued work in this session |
/recover |
Move the next unprocessed prompt into the draft for explicit resend |
/compact |
Summarize and compact conversation history (optional guidance) |
/new |
Start a new session (! stops background tasks) |
/session-id |
Show and copy the current session ID |
/delete |
Delete the current session after confirmation and start a new one |
/fork |
Fork this session and switch to the copy (--worktree [name]: into a new git worktree, to continue in another terminal) |
/rewind |
Rewind to an earlier prompt, or undo the last rewind |
/? |
Show keybindings |
/usage |
Show token usage breakdown |
/version |
Show hob’s version |
/timing |
Show tool timings, or export a snapshot: /timing export PATH |
/queue |
Remove items from queue |
/model |
Switch model |
/account |
Switch the account this session’s provider uses, or add one |
/theme |
Switch color theme |
/mcp |
Configure MCP servers |
/login |
Authenticate with an LLM provider |
/lazygit |
Open LazyGit for this session’s directory (\ hides) |
/cd |
Change working directory |
/btw |
Ask a quick question (no tools, no history pollution) |
/yolo |
Toggle YOLO mode (skip all permission prompts) |
/fast |
Toggle fast mode (Anthropic Opus or Codex subscription models) |
/autoreview |
Toggle session autoreview and verified fixes (base/full/on [level]/off/status/run [level]) |
/workflow |
Toggle workflow mode (task callable inside code_execution) |
/vim |
Toggle vim keys in the chat input |
/paste_collapse |
Show paste-collapse status, set on/off, or set the line threshold |
/youare |
Name the agent in the chat, or show its name |
/hobify |
Toggle the Hob beside each tool call (terminals with Kitty graphics) |
/exit |
Exit the application |
/reload |
Reload plugins and config (! stops background tasks) |
/trust |
Trust this folder and load its shared project config |
/packupdate |
Update packages (++lockfile, ! skips review) |
/packdel |
Remove undeclared packages (++all, or a name) |
/autotitle |
Configure automatic session titles: status, model, endpoint, reasoning, on, off, help |
/cost |
Spend across sessions and projects: /cost [session|rescan|export |
/export |
Export full session transcript across compactions: readable (Markdown, default) or raw (JSON); optional path |
/favorite-model-picker |
Pick one of your favorite models |
/import |
Import a raw JSON session transcript into a new tab; source file is left unchanged |
/infobar |
Show, hide or restyle the info bar: /infobar [on|off|toggle|style [name]] |
/memory |
View, edit, and delete memory files |
/model-picker |
Pick a model, Space or Tab stars a favorite |
/prompts |
Browse this session’s prompts with short summaries |
/rename |
Rename the current session |
/resume |
Browse and switch sessions |
/skills |
Browse skills and insert a $skill-name mention |
/tasks |
Browse and search subagents |
/thinking |
Extended thinking: pick an effort level, or set one directly |
Sessions
Sessions run concurrently. Ctrl+N in /resume starts a fresh session while the current one keeps working in the background, and /resume shows the live status of each (working, needs input, idle) so you can jump between them. When a background session finishes or needs input, Hob flashes a note in the status bar. /rename renames the current session; in the session picker, Ctrl+N / Ctrl+R / Ctrl+D create, rename, and delete. Deleting the current session starts a fresh one and keeps the picker open. /new replaces the current session instead, which ends its turn and background tasks.
Automatic session titles
The bundled autotitle plugin is inactive until configured. Once on, it gives each prompt a key of about five words, and a one-paragraph summary when the prompt has more than 20 lines or 2000 bytes. The title names the main purpose of the session in a few words. The model distills it from the session’s prompt keys and its current title, and it keeps the title while the main purpose stays the same. The model also marks each prompt as a goal, or as a follow-up that only serves earlier work: a review, a fix of review findings, a retry, a commit, or a side question. Once the session has a usable current title, a follow-up-only batch keeps it. A batch can replace that title only if the model identifies a new goal. If there is no usable current title, for example if an older title exceeds the 10-word limit, any batch can supply a replacement. Autoreview prompts that repair findings or failed checks are always follow-ups. A fork keeps its fork of title until the model gives it a title of its own. Manual names and older sessions without title ownership remain protected: their prompts still get keys and summaries, but their titles never change.
/autotitle status
/autotitle model <provider/model>
/autotitle endpoint <chat-completions-url> <model> [api-key-env]
/autotitle on
/autotitle off
/autotitle reasoning <default|none|minimal|low|medium|high|xhigh>
/autotitle helpmodel and endpoint save the backend and turn naming on. Model mode uses Hob’s configured providers and credentials, bills usage to the session, and creates no task tab or conversation messages. Endpoint mode uses a full OpenAI-compatible Chat Completions URL. Supply an environment variable name for optional bearer authentication, never the token itself.
Local/private endpoints also need their exact host and port in net.allowed_private_hosts, followed by /reload. For example, Ollama’s compatible endpoint can be configured with /autotitle endpoint http://127.0.0.1:11434/v1/chat/completions gemma4:e4b. For reasoning models such as Gemma, use /autotitle reasoning none to leave the small output budget for the title, keys, and summaries. default omits the reasoning field for endpoints that do not support it. HTTP redirects are not followed.
Commands persist only the settings they change in autotitle/settings.json under Hob’s state directory. Those keys override plugins.autotitle options. Other Lua settings remain effective after /reload. Initial Lua settings can use active, model, endpoint, api_key_env, reasoning_effort, and timeout. The reasoning command applies only to endpoint mode. Set plugins.autotitle.enabled = false to unload the plugin and its command.
/prompts lists the current session’s prompts by their keys, oldest first, and opens on the newest. Below the list, the selected prompt shows in full, or as its summary when it is long. Enter opens a copy of the full prompt in $VISUAL or $EDITOR, and the browser returns to the same prompt when the editor exits. A session without autotitle data, for example while autotitle is off, lists the transcript’s prompts by their first lines.
The plugin keeps a copy of each prompt it records, with its key and summary, in autotitle/ under Hob’s state directory until the session is deleted. It records nothing while it is off. The first prompt it records in a session also imports the prompts the transcript already holds, so a session started before autotitle was on gets them too. Interactive session deletion removes the copies. Copies left by CLI deletion are removed when the plugin next loads. Disabling the plugin leaves them untouched until it loads again.
Modes and toggles
/yolo: skip permission prompts for this session (deny rules still apply). The toggle survives a resume, and--yoloonly sets the starting value. Config:yolo = true./thinking: extended thinking. Bare, orAlt+T, it opens a picker that shows what each level costs in tokens. With an argument it sets the level directly:off,adaptive,minimaltomax, or a token budget. A model whose API has no adaptive mode does not offeradaptive. A session set to it shows and sends the level the provider uses instead. New sessions start at the last level you set, in-p, SDK and ACP too. Config:always_thinkingpins a level./fast: faster responses on Anthropic Opus, and on eligible Codex models when you sign in with a ChatGPT subscription. OpenAI API keys and every other model ignore it. Config:always_fast = true./workflow: letcode_executioncall thetasktool (and other workflow-only tools) from inside the Python sandbox. Config:always_workflow = true./vim: vim keys in the chat input, for every tab. Hob keeps the choice for the next start. Config:ui.vim_mode = true. See Keybindings for the keys./paste_collapse [on|off|N]: show status, seton/off, or enable collapsing aboveNlines (positive integer). See Paste collapse for behavior and saved settings.- Plan / build: not a slash command. Press
Shift+Tabin the input to toggle plan mode (plan-file writes only). /reload: rebuild plugins and config without leaving the app./btw: one-shot side question with no tools and no history pollution./memory: open the memory file picker (view / edit / delete). See thememorytool under Tools./timing: inspect tool-call summaries, nesting and measured phases for the current/last submission or the whole process. Refresh to include newer activity; durations overlap and are wall-clock, not CPU time./timing export PATHsaves a Chrome Trace snapshot for Perfetto without stopping capture, refuses overwrite, and is user-only. The automatic trace is written on exit; active snapshot spans are incomplete. Local profiling does not require OpenTelemetry.
Custom commands
You can define your own slash commands as Markdown files. Empty files are skipped.
Discovery and priority
Later sources override earlier ones when the command name matches (the stem of the file, or name in frontmatter):
- User config:
~/.config/hob/commands/ - User third-party:
~/.claude/commands/ - Project dirs, walking from the current working directory up to the nearest
.gitroot. At each level:.hob/commands/, then.claude/commands/
Because the walk goes cwd → … → git root, a command at the repository root overrides the same name found only under a nested cwd. Project commands override user commands. Palette names are /project:<name> or /user:<name> depending on which scope won.
Skip all of the above with --no-commands (see CLI).
Metadata
You can add optional metadata at the top of the file between --- lines to set name, description, and argument-hint:
---
description: Review code for issues
argument-hint: <file>
---
Review $ARGUMENTS and suggest improvements.Arguments
Use $ARGUMENTS in the command body. It gets replaced with whatever you type after the command name. The command is treated as accepting args if the body contains $ARGUMENTS or you set argument-hint.
For example, /project:review main.rs replaces $ARGUMENTS with main.rs.
Aliasing commands
Prefer a different name for a command? hob.api.run_command runs any slash command exactly as typing it would, so an alias is a one-line handler in your init.lua instead of a reimplementation.
-- ~/.config/hob/init.lua
local aliases = {
{ name = "/clear", target = "/new", description = "Alias for /new" },
{ name = "/history", target = "/resume", description = "Alias for /resume" },
}
for _, alias in ipairs(aliases) do
hob.api.register_command({
name = alias.name,
description = alias.description,
handler = function()
local ok, err = hob.api.run_command(alias.target)
if not ok then
hob.ui.flash("could not run " .. alias.target .. ": " .. err)
end
end,
})
endAliasing adds a name, so both stay in the palette. Any command works as a target, including MCP prompts and other plugins’ commands. See hob.api.run_command for matching and error handling, or hob.ui.action to bind a key instead of a name.
What a plugin may run
run_command only runs what the calling plugin’s grants cover. Your init.lua counts as a plugin, and its grants come from ~/.config/hob/plugin.toml (see plugin permissions). Commands not listed here run for any plugin. A command sent with hob.ui.input_submit follows the same rules.
- Needs
fs_read,fs_write:/cost,/memory. - Needs
fs_read,fs_write,net,run,env,agent,session_read,session_write,session_settings:/tasks. - Needs
fs_read,fs_write,net,run,env,agent,session_read,session_write,session_settings,input:/skills. - Needs
fs_read,fs_write,net,env,agent,session_read,session_settings:/autotitle,/prompts. - Needs
fs_read,fs_write,run,agent,session_read,session_write,session_settings:/export,/import,/rename,/resume. - Needs
fs_read,fs_write,session_settings:/favorite-model-picker,/model-picker. - Needs
fs_read,run:/infobar. - Needs
agent:/btw. - Needs
agent,session_write:/compact, custom commands, MCP prompts. - Needs
session_write:/new,/fork. - Needs
session_settings:/model,/account,/cd,/fast,/workflow,/thinking. - Needs what its owner holds: every other plugin command, as described below.
- Only when you type it:
/pause,/unpause,/recover,/session-id,/delete,/lazygit,/yolo,/autoreview,/trust,/packupdate,/packdel. No plugin permission grants access to these commands.
A plugin command’s handler runs with the grants of the plugin that registered it. The caller picks the arguments, so it must hold at least the same reach:
- Every permission the owner holds.
- Host access as wide as the owner’s. A caller without
net_hostshas it. A caller with a list needs the owner to have one too, and must contain each of its entries exactly as written, so*.example.comdoes not coverapi.example.com. - Its own provider on the address of each provider the owner registered. That address may be on your LAN, which
net_hostscannot grant.
A caller granted run skips both address checks, since a process it starts can reach any host.
Related: CLI for shell flags and subcommands, Skills for on-demand playbooks.