Slash Commands

Type / in the composer to run a slash command. Commands come from three sources, all surfaced in the same autocomplete dropdown:

• Agent built-ins — handled inside the agent loop.
• TUI commands — handled by the app.
• Skills — bundled, user, and project Agent Skills, invoked as /skill-name.

If a name collides, agent and TUI commands take precedence over a skill of the same name.

Agent built-ins

These operate on the conversation itself.

Command	Description
/help	Show the list of available agent commands
/compress	Compress the message history to reclaim context
/clear	Clear message history; in the TUI, also clears Terminal output and Logs
/reset	Clear message history, Terminal output, and Logs (alias of /clear in the TUI)
/context	Show the current context token count

TUI commands

These control the app and your session.

Command	Description
/skills	List available Agent Skills
/agents	List or validate custom agents (/agents validate)
/init	Create or update AGENTS.md for this repository
/attach	Attach an image: clipboard if no path, or /attach <path> for a file
/detach	Remove pending image attachments
/plan	Switch to Plan mode
/build	Switch to Build mode
/sidebar	Show or hide the side panel
/settings	Open the full-screen Settings editor
/memory	Open Project Memory or run a memory subcommand
/permissions	Show or switch the shell/edit permission mode
/model	Choose the active model
/effort	Choose the active model’s thinking effort
/login	Sign in to a provider, e.g. /login chatgpt
/logout	Sign out of a provider, e.g. /logout chatgpt
/gigacode	Toggle gigacode workflow orchestration on or off
/web-search	Show or set the web tool mode: auto, hosted, client, or off
/goal	Set, show, or clear an autonomous completion goal
/loop	Run a prompt on a repeating schedule
/tasks	Show the shared task list
/queue-clear	Clear queued follow-up messages
/share	Share a live link to this session (/share lan, /share <port>, /share stop)
/copy	Copy the last response to the clipboard
/diagnostics	Show version, model/endpoint, and the local diagnostics log path
/bug	Package local diagnostics into a shareable zip for a bug report
/version	Show the Kolega Code version
/update	Update Kolega Code to the latest version
/quit	Save the session and exit
/exit	Save the session and exit

Sharing a live session

/share starts a server for the session you are in and copies the link to your clipboard. Whoever opens it watches the session as it happens — reasoning, tool calls, terminal output — with a LIVE badge and the option to scrub back through what already happened.

/share             # reachable from this machine
/share lan         # reachable from your local network
/share 9000        # ...on a port you choose
/share lan 9000
/share stop        # stop sharing

Sharing listens on port 8765 unless you name another one. If something already has that port, the share quietly takes a different one and says so; a port you asked for explicitly is treated as a requirement instead, because a tunnel is probably pointed at it.

The link carries a token that is generated per share and is the only thing gating access, so treat it as the secret it is. Sharing stops when you run /share stop or leave the session; nothing keeps listening afterwards.

Viewers are read-only: they cannot type, approve a permission prompt, or interrupt a turn.

Caution

/share lan binds your machine’s address on the local network, so anyone who can reach it there and has the link can read that whole session, including file contents and command output. A live link is also not redacted — unlike kolega-code share export, whatever the agent printed is visible to whoever holds it. Run /share stop when you are done.

To let someone watch from outside your network, leave the share on loopback and forward the port through a tunnel you control. There is no built-in relay. See Letting someone off your machine watch for verified Tailscale, ngrok, and SSH recipes, and for how to turn the link /share gave you into the one you send.

To send someone a recording rather than live access, kolega-code share export writes a static bundle with secrets scrubbed.

Run /goal <condition> to set an autonomous completion goal the agent works toward, verifying its own progress after each turn until the goal is met, the turn cap is hit, or you pause it. /goal (no args) shows the goal status; /goal clear (aliases: stop, off, reset, none, cancel) removes it. See Goal-Conditioned Work for the full loop behavior, safety model, and examples.

Run /loop <interval> <prompt> to re-run a prompt on a schedule — for example /loop 5m check whether CI went green. Cron works too, via /loop --cron "0 9 * * 1-5" <prompt>. Iterations only start when the session is idle, missed windows never pile up, and your typed messages always take priority. /loop status shows the schedule and next fire; /loop stop (aliases: clear, off, cancel, none, reset) or Esc ends it. With no prompt, /loop reads .kolega/loop.md. See Scheduled Loops for the interval and cron syntax, --fresh, caps and expiry, and the unattended-run caveats.

Run /model to open a selectable list of supported models for the current provider. You can also switch directly with /model <name>.

Run /settings to open the categorized Settings editor without navigating to the sidebar tab. Unsaved changes remain a draft until you select Apply Changes.

Run /memory with no arguments (alias: /memory browse) to browse and edit the project’s private memory bank. Lightweight subcommands are also available:

Command	Description
/memory status	Show enabled/backend/identity state, sizes, and the exact bounded startup context the agent receives
/memory on / /memory off	Enable agent access or disable it without deleting data
/memory files	List entries and sizes
/memory show [path]	Show bounded content; defaults to MEMORY.md
/memory path	Show the private local backend directory

See Project Memory for project identity, limits, concurrency, privacy, and model-exposure details.

Run /effort to open a selectable list of supported effort values for the active model. You can also switch directly with /effort <level>.

Run /login chatgpt to sign in with a ChatGPT subscription and use OpenAI models without an API key; /logout chatgpt removes the stored credentials. See Sign in with ChatGPT.

Run /queue-clear to discard follow-up prompts that you queued while the current turn is running. It removes their Queued transcript entries, but it does not cancel or otherwise stop the active agent turn.

Run /tasks to print the shared task list into the transcript. It reads session state directly and never starts an agent turn, so it is available in both interaction modes. Build mode owns the list and is the only mode that can change it; in Plan mode the agent can read it but not modify it.

Run /agents or /agents list to inspect all effective user and project custom-agent definitions, including agents configured for the other interaction mode. Run /agents validate to rescan the files and report invalid definitions. File changes become dispatchable after the active agent is rebuilt (for example, by switching modes) or the TUI is restarted.

Run /diagnostics to print a snapshot of this session — version, platform and terminal, active model and endpoint, which providers have keys, and how many event-loop stalls or LLM errors have been recorded — followed by the path to the local diagnostics log. Run /bug to package that log, any captured stack dumps, and the current session into a single shareable zip for a bug report (API keys are scrubbed; the conversation and file contents are kept). See Diagnostics & Bug Reports for what gets captured, where it lives, and the privacy model.

Run /init to have the agent inspect the repository and create or update a concise root AGENTS.md. Extra text after the command is passed as focus or constraints:

/init focus on Python packaging and test commands

Skills

Any bundled or locally discovered skill is available as /skill-name. Local skills are loaded from user or project .agents/skills/ directories, with project and user versions able to override a bundled skill of the same name. Running a skill loads its instructions (and a manifest of its bundled resources) into the conversation. Pass extra text after the command to run the skill against a specific request:

/release-notes summarize changes since the last tag

Use /skills at any time to see what’s available in the current project.