Slash commands¶
Type /help at the prompt to see this list at runtime.
Basics¶
| Command | Effect |
|---|---|
/help |
List every registered command (built-in + plugin). |
/quit |
Exit letscode. |
/reset |
Clear the conversation history (system prompt stays). |
/model [NAME] |
With no arg on a TTY: inline picker with filter-as-you-type + arrow-key navigation, Enter to select, Esc/Ctrl+C to cancel. Off-TTY: numbered list, type the number on the next line. With an arg: set directly. Interactive picker since v0.7. |
Inspection¶
| Command | Effect |
|---|---|
/tools |
List active tools. |
/skills |
List discovered skills. |
/plugins |
List loaded plugins (built-in + entry-point) with version. |
/reload |
Reload skills from disk and re-read TOML config (since v0.4); honours launch-pinned flags. |
Sessions (since v0.4)¶
| Command | Effect |
|---|---|
/tree |
Show the session as a numbered entry chain. |
/fork [n] |
Branch a new session at entry n (default: all); switch to it, original untouched. |
/clone |
Duplicate the whole session into a new file and continue on the copy. |
/new |
Start a fresh session in the same project directory. Since v0.7. |
/resume [N] |
Without an arg, list prior sessions with labels (newest first). With N, switch to session #N. Since v0.7. |
/name [TEXT] |
Show the current session's human label; with an arg, set it. Auto-set from the first user message unless already labeled. Since v0.7. |
Session output (since v0.7)¶
| Command | Effect |
|---|---|
/copy |
Copy the last assistant text to the clipboard (pbcopy / wl-copy / xclip / xsel / clip.exe, whichever is on PATH). |
/export [PATH] |
Write a self-contained HTML transcript. Default path: ./<session-label>.html in the cwd. |
/share |
Upload the HTML transcript as a secret GitHub gist via gh gist create (requires gh on PATH and authenticated). |
Skills¶
/skill:<name> [message]: splices a skill body into the next user message.
Equivalent to typing the skill body followed by your message. The model also has the skill exposed as a tool named skill_<normalised-name> (auto-wrapping), so it can invoke skills natively via tool-calling without needing the slash form.
Skills covers the format and discovery paths.
Steering and follow-up¶
These mid-flight UX commands let you queue work without aborting the current turn.
| Command | Effect |
|---|---|
/steer <text> |
Queue a steering message. Delivered at the next tool-batch boundary. |
/follow-up <text> |
Queue a follow-up. Delivered after the agent would otherwise emit agent_end. |
/queue |
Show pending steering and follow-up messages. |
Streaming-time steering
Since v0.4 you can type while the model is working: Enter stops the run; type then Enter stops and redirects it to the new instruction (interrupt-and-redirect). /steer and /follow-up remain the explicit queue commands and back the RPC steer/follow_up mechanism.
Compaction¶
/compact [bias]: manually compact older history into a summary.
When the context window is approached, the agent auto-compacts. Manual compaction is for taking the bias parameter:
Without a bias, the default summariser prompt runs.
UI toggles¶
These are frontend-owned commands. They're registered by BasicFrontend.__init__, not via the standard letscode_register_commands hook, because they toggle frontend-private state.
| Command | Effect |
|---|---|
/verbose |
Toggle full vs. truncated tool-result panels and tool-call args. Default: truncated. |
/footer |
Toggle the per-turn model/cost/ctx footer. Default: on. |
In default mode, the read tool's result panel is suppressed entirely. Only the call line > read({"path":"X"}) shows. /verbose restores the full syntax-highlighted body.
Plugin commands¶
Slash commands from installed plugins appear in /help alongside the built-ins. See the reference plugin for /remember, /memories, and /forget <id> (from letscode-memory).