Getting started and conventions
See the quickstart for installation and prerequisites. lurus --help, lurus --version and lurus <command> --help show local help and version information. There is no separate help command.
D means a direct action, P an agent task and D+P a direct pre-action followed by a task. Direct actions can incur costs, access the network and change files. Replace placeholders with actual values; do not translate command syntax.
See diagnostics for proxies and certificates, and settings and permissions for configuration.
Authentication and management
| Command | Options | Effect |
|---|---|---|
lurus login | --api-key, --email, neither takes a value. | Browser login by default, with alternative interactive input. If both flags are supplied, API-key input is used. Do not pass credentials as arguments. |
lurus register | None. | Registers through the browser and saves credentials. |
lurus logout | --all | Removes stored authentication and attempts server logout for one or all JWT sessions. Does not revoke API keys or remove environment variables. |
lurus status | None. | Shows the authentication source, profile and proxy information. |
lurus doctor | None. | Checks the environment without requiring login. |
lurus update | --check | With --check, checks the version and may write the cache. Otherwise installs a newer version globally through npm; development builds do not install. |
Handled errors and cancellations do not always produce a nonzero exit code. Inspect output; success text does not guarantee server-side revocation. See troubleshooting.
Chat and single prompts
chat is the default command. For example: lurus chat --mode ask --prompt "Explain the project".
Danger: dangerouslySkipAllPermissions disables permission checks without the interactive phrase dialog used by /trust on, unless globally blocked. Do not use it as a default.
| Option | Values and effect |
|---|---|
-m, --model <model> | Public model selection or alias. |
-p, --prompt <prompt> | Runs a single prompt. lurus chat [prompt...] is also supported. |
--max-turns <n> | Positive safe integer in decimal digits. Limits main-agent model requests including retries. Single prompts only, not interactive chat or --pipe; invalid combinations exit with code 1 before authentication. |
-c, --continue | Loads the latest session. |
-r, --resume <id> | Loads the specified session. |
--pipe | Starts a long-lived stdin/stdout integration. |
--output-format <format> | Defaults to text; also accepts stream-json or json. Both JSON variants produce NDJSON, not a single final object. |
--add-dir <dirs...> | Adds directories; permission rules still apply. |
--from-pr <number> | Tries git diff origin/main...HEAD first and gh pr diff <number> only on failure. Limited to 50,000 characters; not guaranteed to load that PR. |
--json-schema <path> | Adds schema instructions to the prompt; no automatic response validation. Read failures warn and continue. |
--permission-mode <mode> | default, acceptAllEdits or dangerouslySkipAllPermissions. Separate from interaction mode. |
--thinking <mode> | on, adaptive or off, subject to model support. |
--effort <level> | minimal, low, medium, high, xhigh or max; checked before authentication. |
--max-mode <on|off> | Explicitly sets Max mode. |
--mode <mode> | agent, ask or plan. |
--debug | Starts debug interaction mode. |
Single prompts save the session and exit with code 0 or 1. For machine output, use --output-format json with a separate value; short and = forms do not activate the same log redirection. See sessions and automation.
Batch
Warning: Batch can change files and execute processes. A request to read only does not enforce read-only access.
lurus batch prompts.txt --output-format json --continue-on-error
Options are -m/--model <model>, -o/--output-format <text|json>, --continue-on-error and --max-turns <n>. Output defaults to text and execution stops at the first error. The turn limit bounds model requests including retries per prompt, using the same integer format as above. A separate --- line divides prompt blocks; individual lines are not automatically separate prompts. History is cleared between prompts. Failures result in exit code 1 even with continuation; an empty list exits with 0. See batch files and JSON output.
MCP commands
| Syntax | Options and effect |
|---|---|
lurus mcp add <name> | Exactly one of --command <cmd> or --url <url>. Other options: --type <stdio|http|sse>, --args <args...>, --env <pairs...>, --cwd <path>, --header <pairs...>, --timeout <ms>, --trust, --project. Global without the project flag; remote type defaults to http. Timeout requires positive integer milliseconds. Saves but does not start. |
lurus mcp remove <name> | Optional --project; removes configuration and header references. Unknown entries or errors exit with code 1. |
lurus mcp list | Lists global and project servers, with no options of its own. |
add can overwrite same-name configuration. Review existing entries and scripts first. Environment, arguments and working directory are local-only, headers remote-only; pairs use KEY=VALUE. Example: lurus mcp add local-docs --command node --args /absolute/path/server.mjs --project. See MCP.
CI/CD and hidden integration
These are the public CI commands. Reports may overwrite files. Scans can incur costs and transmit code; PR actions are externally visible.
| Command | Flags |
|---|---|
lurus security-ci | -m/--model <model>, -f/--format <sarif|json|html|text> (default sarif), -o/--output <path> (default lurus-security-results.sarif), --fail-on <severity> (default high), --diff, --diff-base <ref> (help lists HEAD), --no-upload, --pr-comments, --comment-min-severity <severity> (default medium). |
lurus code-review-ci | -m/--model <model>, -f/--format <json|html|text> (default json), -o/--output <path> (default lurus-code-review-results.json), --fail-on <severity> (default high), --full, --staged, --diff-base <ref>, --pr-comments, --verdict, --comment-min-severity <severity> (default medium). |
Severity levels are critical, high, medium, low and info; invalid values fall back to high. Exit codes: 0 for a passed check, 1 for blocking findings or missing entitlement, 2 for execution errors and 3 for exhausted balance. --no-upload prevents findings upload, not model requests. See CI/CD for examples and boundaries.
Every slash command
Sessions and tasks
Warning: /sessions delete <id> deletes without another dialog. /build-plan requests implementation, not a preview. Check IDs and plans first.
| Syntax | Effect |
|---|---|
/resume [search text...] | D: Lists up to 20 project sessions, does not load them. |
/sessions [all|delete <id>] | D: Lists up to 50 project or cross-project sessions. Deletes with delete; without an ID, shows the list. |
/recall <search text...> / /recall --load <id> | D: Searches or loads across projects. Another project requires confirmation, defaulting to No. |
/clear | D: Clears chat, not the task or files. |
/task new / /task status | D: Starts a new task or shows its status. |
/build-plan <path...> | D+P: Activates the plan and requests implementation in Agent mode. |
/save [path] | D: Exports Markdown, with a timestamped filename by default. External paths and overwrite require consent. |
/rewind chat | D: Rewinds messages, not file changes. |
/quit, /exit | D: Ends the session. |
See chat and sessions.
Models and modes
| Syntax | Effect |
|---|---|
/model [name/alias] | D: Lists public models or selects one immediately and globally. No clear or reset. |
/utilitymodel [name/alias|clear] | D: Changes the global utility model or removes the override; restart required. |
/compactionmodel [name/alias|clear] | D: Changes global compaction selection; restart required. |
/mode [agent|plan|ask|debug] | D: Shows or changes session mode; defaults to agent. |
/agent [task...], /plan [task...], /ask [task...], /debug [task...] | D/D+P: Switches mode without text and also submits a task with text. Debug does not automatically start diagnostic recording. |
/max | D: Toggles and saves globally; appending on or off does not explicitly set it. |
/thinking [on|adaptive|off] | D: Current and global setting, default on. Always-on thinking cannot be disabled. |
/effort [minimal|low|medium|high|xhigh|max] | D: Default high; model-dependent support. |
/costsaving [on|off] | D: Shows status or explicitly sets it; enabled by default, not a toggle without a value. |
/displaycost [on|off] | D: Current and global display, off by default; cost tracking remains independent. |
See models and costs and settings.
Context and configuration
Warning: /rules create can overwrite files. Repair resets damaged global consent data for all projects, with backup and confirmation. /trust on disables checks and requires an exact risk confirmation; prefer narrow approval.
| Syntax | Effect |
|---|---|
/context | D: Shows token breakdown. |
/copy | D: Copies the last agent response to the clipboard. |
/paste | D: Saves and attaches a clipboard image without a prompt. |
/detach [index|name fragment], alias /drop | D: Removes references, not files. No argument removes all. |
/compress, alias /compact | D: Compacts context; detail loss is possible. |
/refresh, alias /reindex | D: Refreshes the @ file-search cache, not the semantic index. |
/indexing [on|off|clear|delete|rebuild] | D: Shows status, changes the current setting or deletes/rebuilds the index. Project values do not reliably survive restart. |
/config | D: Shows loaded configuration; no set, get, reset or reload. |
/init | D+P: Creates a starting structure and requests analysis and completion. |
/rules [list|show <name>|create [name]] | D: Lists, shows or creates project rules. Default name new-rule; restart after changes. |
/permissions | D: Shows sources, IDs and scopes. |
/permissions revoke <id> / /permissions clear | D: Removes persistent project grants. |
/permissions revoke session <id> / /permissions revoke settings <id> / /permissions revoke global-settings | D: Revokes the respective source. Settings JSON remains. |
/permissions repair / /permissions repair trust | D: Repairs damaged global permission or consent data; confirmation defaults to No. |
/trust [on|off] | D: Shows status, disables checks with on or restores them with off. Can be blocked globally. |
See context, rules, settings and permissions.
Extensions
Warning: Generators and updates can overwrite files; /memory clear deletes without another dialog. Review sources, names and destinations first.
| Syntax | Effect |
|---|---|
/commands | D: Shows loaded templates, without management or reload subcommands. |
/agents, /agents list | D: Lists custom agents. |
/agents show <key> | D: Shows metadata and path, not the full prompt. |
/agents trust <key>, /agents untrust <key> | D: Saves or removes trust for the source and file contents. |
/agents run <key> <prompt...> | D: Starts a custom agent; trust is checked and partial results are possible. |
/skills, /skills show <key> | D: Reloads skills and shows the list or contents. |
/skills activate <key>, /skills deactivate <key> | D: Persistent source-bound activation or bundled skill blocking. |
/skills activate-session <key>, /skills deactivate-session <key> | D: Temporary change; persistent blocks remain. |
/skills audit | D: Checks short contents and duplicate names, not security or reload. |
/skillify | P: Requests questions, review and creation of useful skills. |
/create-hook <event> <name> [--ts] | D: Generates project files; TypeScript requires npx tsx. |
/hooks validate | D: Checks part of project configuration. |
/mcp, /mcp restart <server>, /mcp refresh-tools | D: Shows status, reconnects or refreshes tool definitions. Restart alone does not update them. |
/memory, /memory clear, /memory edit, /memory sweep | D: Shows, deletes, edits or consolidates memory. |
/install, /install list, /install remove <package>, /install update [package], /install [bundle] <sources...> | D/D+P: Help, package list, confirmed removal or installation with a copy task. Updating without a name is only an agent task. |
See custom commands, agents, skills, hooks, MCP and memory.
Tools, analyses and Git
Warning: Undo restores older contents in agent-affected paths without another dialog; your later edits there can be affected. Check /undo list and protect your changes first. A read-only request does not authorize Git or PR actions.
| Syntax | Effect |
|---|---|
/tools [desc] | D: Lists tools by mode and plan, does not toggle them. |
/web <url> | D+P: Fetches up to 30,000 characters and requests a summary. |
/test gen|generate [target] | P: Pro or above; write, run and fix tests, targeting the project by default. |
/docs [target] | P: Pro or above; documentation, targeting the project by default. |
/tdd on|warn|strict|off, /tdd <feature> | D/D+P: on means warn; feature text enables strict and requests Red/Green/Refactor. |
/tdd-implement | P: Pass existing tests, refactor and commit; does not itself enable a guard. |
/edit | D+P: Opens the prompt editor; no /editor alias. |
/diff | D: Shows checkpoint diff with a 60-line preview. |
/undo [list|all|N] | D: 1 is the newest and all the oldest retained project checkpoint. Agent-affected paths only; history is truncated. Without arguments and with one checkpoint, restores directly. |
/commit [message] | D: Optional staging approval and commit dialog, defaulting to cancellation; no push. |
/create-pr [--draft] [--reviewer <value>] [--label <value>] | D+P: Checks GitHub prerequisites and requests externally visible PR creation. |
/fix-issue <number> | D+P: Fetches the issue, attempts branch creation and requests a fix. |
/analyze-issue <number> | D+P: Requests read-only analysis but does not enforce a mode. |
/fix-pr | D+P: Fetches current PR comments and requests fixes; no push. |
/security-review, alias /scan | D: Native security analysis on Pro or above. |
/review, alias /code-review | D: Native code review on Pro or above. |
Shared analysis flags: --html, --json <path>, --model <selection>, --tier <max|eu|custom> and --provider <eu|anthropic|openai|openai-global|gemini|global|lurus>. Security also supports --diff, --diff-base <ref> and --focus <crypto|auth|injection|supply-chain>. Review supports --full, --staged, --diff-base <ref> and --no-prechecks. See workflows.
Media, orchestration and information
Warning: Media and orchestration actions may incur costs and write files. Reports overwrite same-name files without another dialog; HTML loads an external chart library. Review contents before sharing.
| Syntax | Effect |
|---|---|
/image [--model=<id>] [--aspect=<value>|--size=<WxH>] <prompt>, alias /img | D: Default size 1024 × 1024 pixels. /image models [--refresh] lists models. |
/image-edit [options] <path> <prompt>, alias /img-edit; alternatively repeated --source=<path> | D: 1–5 ordered sources, new result file. |
/video --seconds=<4|8|12> [--portrait|--landscape] [--ref=<path>] <prompt>, alias /vid | D: Duration required, portrait by default, MP4 output. |
/orchestrate [options] <prompt> | D: ULTRA required; cumulative budget defaults to 20 EUR, minimum 5 EUR. |
/orchestrate status <id>, /orchestrate resume <id>, /orchestrate abort <id>, /orchestrate budget <id> <total> | D: Run control. Budget changes require a pending budget decision, then explicit resume. Abort does not guarantee file rollback. |
/ide, /ide enable, /ide disable | D: Shows status, connects or disconnects the IDE. |
/help | D: Static help. |
/cost | D: Cost, tokens and cache usage. |
/stats | D: Duration, model, turns and tool definitions. |
/doctor | D: Environment diagnostics. |
/status | D: Session, budget, context and proxy information. |
/changelog | D+P: Git comparison and changelog task. |
/report [--json [filename]] | D: Defaults to lurus-report.html, alternatively JSON. Filename only, not an arbitrary target directory. Metadata and counters, not a transcript. |
/mermaid [target] | P: Diagram code, targeting project architecture by default. |
Media options use =; see media for supported sizes and formats. Orchestrator flags: --mode, --max-cost, --model, --planner-model/--analyst-model, --implementer-model/--architect-model, --reviewer-model/--critic-model. There is no cancel, pause or start subcommand. See orchestration and IDE and diagnostics.
Practical entry points
Start with the quickstart, configure settings and use agents for scoped reviews. Workflows covers tests and Git, and automation machine use. Review permissions and distinguish direct actions from agent tasks. Custom extensions can provide additional commands.
Example workflows
See written examples for login and first chat, batch and security and code review.