Start chat and choose a mode
lurus
lurus chat --mode ask
lurus chat --mode plan
lurus chat --debug
lurus chat --continue
lurus chat --resume <session-id>
chat is the default command. --continue loads the latest session; --resume loads a session by ID. Replace placeholders with actual IDs. --mode accepts agent, ask and plan; use --debug for debug mode. In chat, use:
/mode
/mode ask
/ask Explain the current architecture without edits.
/plan Add input validation and define acceptance tests.
/agent Implement the approved change.
/debug Investigate the failing test.
/mode reports/changes the session mode, not a global persistent setting. Shortcuts without text switch modes; with text they switch and submit a task. Debug shortcut does not start a debug session automatically.
| Mode | Boundary |
|---|---|
| agent | Normal implementation with policy-controlled edits. |
| ask | Read/answer; ordinary workspace edits blocked. |
| plan | Research and bound plan preparation, not ordinary implementation. |
| debug | Implementation plus explicitly started debug session/logs. |
Shell commands and processes are also available in Ask and Plan, subject to permission checks. “Read-only” does not mean that no processes can run. Custom agents and subagents may have additional restrictions. A prompt or mode label alone does not grant write access.
Task versus conversation
/task status
/task new
/clear
/rewind chat
/task status shows the task status. /task new starts a new task with a fresh task list and progress record. /clear clears the chat but leaves files and the current task unchanged. /rewind chat removes recent messages, not file changes. Use diff and undo for file checkpoints.
The task record preserves decisions, progress, blockers and verification results. Memory instead preserves knowledge for future tasks. Explicitly decide whether to continue or cancel an active plan before starting a new task.
Create and implement a plan
- Discuss the task, constraints and acceptance criteria in Plan mode.
- Answer questions and review the written plan; no implementation yet.
- Save it in the workspace, for example
.lurus/plans/input-validation.md. - Explicitly request implementation later or select the reviewed plan with
/build-plan. This command requests implementation, not a preview:
/build-plan .lurus/plans/input-validation.md
/build-plan switches to Agent mode and requests implementation. Use a valid plan file generated by Lurus; arbitrary Markdown is not sufficient. The agent completes the task list and records the checks it performs. Explicitly request changes, replacement or cancellation of an active plan.
Find and load sessions
/resume validation
/sessions
/sessions all
/recall validation
/recall --load <session-id>
| Command | Actual action |
|---|---|
/resume [search text...] | Lists up to 20 project sessions by title, path or model; does not load them. |
/sessions | Lists up to 50 project sessions. |
/sessions all | Lists up to 50 sessions across projects. |
/recall <search text...> | Searches across projects. |
/recall --load <id> | Loads history and model. Loading another project requires confirmation, defaulting to No. |
/sessions delete <id> | Deletes the stored session without another confirmation dialog. A deletion lock may prevent the operation. Without an ID, it displays the list. |
CLI --resume and slash /resume perform different actions. Use IDs, not display titles, to load or delete a specific session.
Warning: The following action deletes the stored session without another dialog. Check the ID carefully; recovery is not guaranteed.
/sessions delete <session-id>
Export and exit
Check the destination path and existing files before exporting. The text may contain sensitive content; review it before sharing.
/save notes/session.md
/quit
/save [path] exports the conversation as Markdown. Without a path, it uses a timestamped filename. Absolute paths, destinations outside the project and existing files require consent. Attachments are not exported as complete files.
/quit and /exit end the session. They neither delete the stored session nor undo file changes. Automation covers single prompts, output formats and pipe mode.
Models and session cost
/model lists the public model selection. An argument changes the model immediately and saves the selection globally. /thinking on|adaptive|off and /effort minimal|low|medium|high|xhigh|max also change the current and global settings, subject to model support.
/max toggles Max mode; appending on or off does not explicitly set its state. /costsaving shows the status; on or off sets it explicitly. Cost saving is enabled by default at startup. /displaycost defaults to off and affects display, not cost tracking. Changes to utility and compaction models are global and require a restart. See models and costs and settings.