#Channel slash commands

Commands control the session in the channel conversation where you send them. Start with /help. In a group, address the bot: @clisbot /status. In a DM, /status works. Commands and their replies stay in that conversation; other participants can see them.

A Route and its Rules decide who can chat. Chatting with its Agent gives you no access to the Host, Project or Clisbot app. Managed access supplies the personal permissions needed to choose a different location or configuration.

#Choose a Project and Workspace

A Host connects your app to the daemon that runs agents and holds files. A Project is a registered directory on that Host. A Workspace groups sessions working in one directory; it can use the existing checkout or a separate Git worktree. See Workspaces.

These commands need a direct Route already serving this conversation. They work before its first session starts; mentioning the bot cannot create a Route for an unserved topic.

You want	Send
Current session and next session's location	/project or /workspace
Projects you can use on this bot's Host	/project list
Find a Project	/project search payments --page 2
Choose a Project; clear the Workspace choice	/project <project-id>
Registered Workspaces in the selected/default Project	/workspace list
Only worktree Workspaces	/workspace list --worktree
Find by name, ID, branch or final folder	/workspace search login --worktree --page 2
Inspect without selecting	/workspace show <workspace-id>
Choose an existing Workspace	/workspace <workspace-id>
Return to the Route's Project and Workspace defaults	/project clear
Keep the Project and clear only the Workspace choice	/workspace clear

IDs and unique exact names work for selections and Workspace show. Use IDs when names repeat. Project and Workspace searches match the final folder name, not parent folders. Each list/search returns up to 10 results with counts and next/previous commands; add --page 2 when another page exists. Workspaces show recent activity first. Projects show name order. Inaccessible Projects are excluded before searching and counting.

Bare /project, /workspace and /branch show Current session and Next session, including names, IDs and the Host. Workspace details include branch and full path. They show an uncreated Workspace as new, without inventing a title or worktree path.

#Create the Workspace before starting an Agent

/project list
/project <project-id>
/branch list
/workspace new main
/new fix the flaky login test

/workspace new main creates and selects a worktree Workspace immediately. The Agent starts with /new <message>; bare /new clears the binding so the next accepted prompt starts it. For refreshed remote history, use /branch fetch then /workspace new origin/main.

/workspace new origin/main --branch fix/login --name "Fix login"
/workspace new --local --name "Main checkout"

--branch preserves the branch name you specify; omit it for automatic naming. --name sets the Workspace title. Omit the base ref to use the Host's default branch choice. --local uses the Project's existing checkout and accepts --name, but cannot be combined with a base or --branch. A non-Git Project needs --local.

Selection changes the next session, leaving the running session in its current directory. The choice is shared by everyone in the Route's binding scope (thread, topic, or whole conversation) and persists across /new and Hub restarts.

• /new and /quick use the next-session choice; /quick leaves the current binding in place.
• /fork carries the current transcript and continues here. It uses an explicitly selected destination Workspace; when the destination Project changes without one, it creates there. Otherwise it continues in the source Workspace.
• /side stays in the source Workspace.
• /resume checks the existing session's actual Project and updates the next choice after attachment.

With no explicit choice, /new and /quick normally get a new Workspace named from the prompt, while /fork and /side reuse the source Workspace. The operator can disable automatic organization. If a choice expires after a Route or Host change, reselect or use /project clear; the refused prompt is not automatically retried.

#Location permissions

All location commands need project.use. Choosing or creating a destination also checks the following permissions and configuration grants:

Action	Required access on the destination Project
Discover/inspect Projects, Workspaces, branches and worktrees	project.use
Select or clear a location	agent.create and an allowed Agent configuration
Create either kind of Workspace	workspace.create, agent.create and an allowed configuration
Refresh a Git remote	approval.command

An inherited Route setting that creates a worktree also needs workspace.create in an explicitly selected Project. The Host must support Workspace commands and Git commands; an older Host asks you to update. An operator can disable the whole location-command surface for a Route with workspace.commands: false.

#Inspect Git and refresh branches

Use /workspace for creation and selection. /worktree inspects Git checkouts, including those that are not registered as Workspaces.

You want	Send
Current/next branch and location	/branch
Local and remote-tracking refs	/branch list
Search refs, with pagination	/branch search origin/ --page 2
Refresh origin	/branch fetch
Refresh another configured remote	/branch fetch upstream
Inspect this session's worktree	/worktree
All Git worktrees, including the primary checkout	/worktree list
Find by branch, final folder, Workspace name or ID	/worktree search login --page 2
Inspect without selecting	/worktree show <path-or-workspace-id>

Branch and worktree lists show latest commit first, up to 10 per page. Branch refs retain their namespaces (main, origin/main, upstream/main). They include short hash, commit date in UTC and author. Lists use refs already on the Host; they never fetch implicitly.

Fetch uses the Host's Git credentials and updates remote branch refs. It preserves local branches, tags, HEAD and working files. It needs command approval permission, even though listing does not.

Worktree replies include Host, Project, branch, full path, associated Workspace, file-status counts and latest commit; show includes the commit subject. Counts exclude ignored files and reveal no filenames or diffs. Dirty files do not change list ordering. Before a session, bare /worktree inspects the chosen/default Workspace or the Project root with no Workspace selected. Replies distinguish local (Project root) from worktree. When several Workspaces share a checkout, list/show by path reports their count; show <workspace-id> identifies one association.

#Start, stop and guide sessions

You want	Send
Fresh session; next prompt starts it	/new (/reset)
Fresh session with a first prompt	/new fix the login test
Continue an existing session here	/resume <agent-id>
Copy the transcript and continue here	/fork [message]
One-off with the transcript	/side <message>
One-off without context	/quick <message>
Stop the turn or active Automation runs	/stop (/cancel)
Nudge a running turn	/steer <message>
Send after the current turn ends	/queue <message>
Open the same session in the app	/cowork (/open, /app)

/resume needs agent.create and access to the target session's Project and configuration; a session bound to another conversation cannot be resumed here. Forking needs a Host that supports copying session context. One-off sessions send their answer here and auto-archive. Queued messages run in order, one per turn; detaching the session or restarting the Hub clears them.

On an Automation Route, /stop, /status, /me, /help, /cowork, /followup and approval replies apply. Location, session creation, configuration and extension commands require a directly bound Agent; the bot reports when a command is unavailable.

/cowork needs agent.interact. It returns browser and installed-app links when your operator configured the app web origin; otherwise it returns a copyable clisbot:// URL. Opening a link still needs your own Host access.

#Choose an Agent configuration

Choice	Send
Agent profile (provider, model, mode and thinking)	/agent list, /agent search review, /agent <name>
Provider	/provider list, /provider search codex, /provider <id>
This provider's models	/model list, /model search sonnet, /model <id>
This model's effort levels	/effort list, /effort <id> (/thinking)
Provider's permission mode	/permission list, /permission <mode> (/mode)

A bare configuration command shows the current and next-session choices without changing them. Add list or search <text> to discover options. Choices are bounded by your agent.interact access and provider/model/thinking grants. Models and effort levels belong to their current provider and model; a cross-provider model choice is refused.

Same-provider model, effort and mode changes apply live. A different provider is staged while ordinary messages continue to the running session. Use /new to start clean or /fork to carry context into that provider. While a provider is staged, configuration commands edit that next choice. Unattended modes need the matching approval.* privilege. Fast mode also needs agent.fast.use, including through a profile or resumed session.

/routedefault shows the serving Route and its default. /promoteroutedefault makes this conversation's configuration the default for future sessions on that Route; /promoteroutedefault undo restores its previous default. Promotion needs channel.manage (Connection Manage access); running sessions keep their setup.

#Skills, shortcuts and follow-up

Use /skill list, /skill search <text> or /skill <name> to discover or run an Agent skill. /command list, /command search <text> and /command <name> discover or run account-shared shortcuts.

/command add standup summarize today's changes and blockers
/standup
/command remove standup

Adding/removing shortcuts needs approval.config. Platform command names are reserved; /skill <name> explicitly invokes a skill even when its name collides. Unknown slash words pass through as Agent prompts when addressed to the bot.

/followup shows whether plain messages need a mention in this conversation. /followup auto allows unmentioned follow-up for 5 minutes after the Agent's latest activity, unless your operator set another duration. /followup mention-only requires a mention each time; /followup pause pauses until the next accepted mention; /followup resume restores the Route's setting. In Slack, send it inside the thread you want to change.

/followup route shows the Route's setting. /followup route auto 10 or /followup route mention-only changes it for all conversations and needs channel.manage. Conversation overrides persist until /followup resume; the active follow-up window is lost on Hub restart.

#Identity and approvals

/status (/state) shows chat/thread/topic IDs and whether you can talk here. An admitted, linked Hub Member also sees session details. /me shows your sender ID and linked Member's access. In Telegram and other channels that hide group members' IDs, send /me in a DM. Give these IDs to an admin when asking for access. Use /link <code> from the Managed access identity-linking flow to connect your channel identity to your Hub Member.

/help, /me and /status answer anyone by default, including senders a Rule refuses. The operator can restrict them to admitted senders/managers. Other chat controls need Route admission; configuration, location and approval commands additionally check the permissions described above. An unlinked identity uses the Guest group's grants, which are empty by default.

/approve allows the newest pending permission request; /approve <id> targets one. /deny [id] refuses it. Answer an Agent question with /approve <id> use the staging database. Permission approvals need the open prompt's authority; questions can be answered by participants the Route admits. Buttons work where supported; WhatsApp also uses reactions.

If an acknowledgement is lost or the bot reports an interrupted command, inspect /status or the app before sending a new command message: the first may already have changed the session.

#Channel invocation

Send one complete command message. A command inside surrounding prose is an ordinary Agent prompt. /, backslash and bare verbs share the parser, but channel clients can intercept slash text.

Channel	Invocation
Slack	/clisbot status when the native umbrella is configured; in a channel use @clisbot status or @clisbot \status to avoid native-name collisions.
Telegram	DM: /status. Group: /status@yourbot, @yourbot /status, or the mobile glued @yourbot/status. Bare group /status names no bot and is ignored.
Discord	Native /clisbot with its command text option set to status; acknowledgement is private, command output goes to the conversation. Or mention the bot with command text.
Google Chat	/clisbot status after operator registration in the app console; otherwise command text, mentioning the app in a space.
Feishu / Lark	Type command text and mention the bot in a group. @everyone/@all does not address it.
Zalo Official Bot	Type /status in the direct chat; no native slash menu.
Zalo Personal, WhatsApp	Type command text; address the bot/account in a group.

Several bots can share a room, so outside a DM a command must name this bot even when ordinary messages have a follow-up window. Typed approval answers are exempt: they target a prompt the bot posted and stay silent when it has none. Platform-specific transports and live evidence remain on each channel's page; shared syntax does not establish live verification.