OKLMdocumentation v1.0.3
On this page

OKLM agent commands

Generated by oklm agent commands --markdown from the command registry; do not edit by hand.

Status: generated.

Built-in commands

Command Aliases Arguments What it does
/clear reset new Start a new conversation
/compact [focus] Summarize the conversation to free the context
/context Show what fills the context
/rewind checkpoint undo Go back to before a message: the conversation, the files, or both
/export [file] Write the conversation as Markdown
/mode [mode] Choose the permission mode
/plan Switch to the plan mode
/add-dir Add a working directory for the session
/output-style [style] Choose the output style
/theme [theme] Choose the colors
/config settings [get|set|unset KEY [VALUE]] Show or change the settings
/permissions allowed-tools [allow|ask|deny RULE] Show or change the permission rules
/init Write AGENTS.md for the project
/memory [N] Show the memory files
/tasks jobs bashes [stop N | output N | reap] List the background jobs and agents; stop one, show what it wrote, stop what earlier sessions left
/thinking Switch the full view (reasoning live) and the classic one
/verbose Show tool output whole, or shortened
/tui Enter or leave the full-screen interface
/vim vi [on|off] Vi or Emacs line-editing keys
/help List the commands
/skills [NAME [on|menu-only|model-only|off]] List the skills, show one, or change one's visibility
/reload-skills Read the skills and commands again, telling what changed
/agents [new NAME DESCRIPTION|edit NAME|reload] List the sub-agent types; write a new one; reload them
/hooks [reload] Show the hooks; reload them after a change
/model [N|PATH|NAME] [s] Choose a model in the list, or switch to the one named (saved as the default unless s)
/context-size ctx [TOKENS [force]] Show the context's size and what the model takes, or change it (the conversation is kept)
/effort [low|medium|high|max|auto|status] How much the model reasons before answering
/sandbox [on|off|network on|network off] Commands confined or not, the network allowed in it or not, for this session
/status The version, the model, the settings' files, the session and what runs
/usage cost stats [stats] This session's tokens and time; with stats, the project's sessions and tools
/bug share [report] Write a local bug report bundle: the report, diagnostics, and as much of the conversation as you choose
/feedback [report] Write a local feedback bundle, as /bug does
/doctor checkup [fix|prompt-audit] What is wrong in the installation and the settings; fix it; or have the instructions reviewed
/diff [stage|unstage|revert|open FILE] The work tree's changes, the agent's marked; stage, unstage, revert or open a file
/cd Move the session to another directory, its conversation kept
/import [claude|agent] [--dry-run] Copy another agent's commands, agents and skills into .oklm
/skill-doctor Each skill's cost in the prompt and its uses in this project's sessions
/statusline [command|clear] The command whose first line is the status line
/color [color|default] The prompt's color for this session
/todos Show the task list
/btw [question] Ask a side question, kept out of the conversation; the past ones without a question
/copy [N] [file] Copy the Nth latest answer to the clipboard, or write it to a file
/recap The session in a sentence
/rename [name] Name this session (the model names it without a name)
/branch [name] Go on in a copy of this conversation; the original is kept for /resume
/resume continue [session] Switch to another session, picked or named
/autocompact [auto|TOKENS] When the conversation is compacted by itself
/plugin plugins The plugins installed: their contents, on or off
/services [start|stop|restart|logs NAME] The project's declared services: their state; start, stop, restart, see the logs
/mcp [tools|reconnect|enable|disable|log|login|logout NAME] The MCP servers: their state and tools; reconnect, turn on or off, see the log, sign in or out
/background bg [request] This conversation goes on in the background on its own; a new one starts here
/fork request A copy of this conversation works on a request in the background
/schedule routines [WHEN PROMPT | remove ID | run ID | log ID] Tasks the crontab runs as oklm agent -p; list, add, remove, run now, see the last run
/loop proactive [interval] [prompt] | stop Run a prompt again and again: on an interval, or paced by the model
/goal [condition|clear] Keep working across turns until a condition is met
/exit quit Leave

Built-in prompt commands and skills

Prompt commands (SKILL.md text shipped with OKLM; a project or user skill of the same name replaces one).

Command Arguments Who starts it What it does
/skill-creator [what the skill should do] the user or the model Write a new skill for this agent, or improve one. Use when the user wants to create, change, check or test a skill (a SKILL.md of instructions loaded when a task needs them), or to turn a workflow they repeat into one.
/commit [what to say about the change] the user Commit the changes of this session, the way this repository writes its commits.
/pr [base branch] [--draft] [what to say] the user Open a pull request (or GitLab merge request) for this branch, titled and written from all of its commits.
/code-review [low|medium|high|max] [--fix] [--comment] [--max-findings N|all] [PR number|branch|path] the user Review the current changes, a pull request, a branch or a path for bugs: findings ranked by severity, each with its file and line, why it is wrong and a fix.
/security-review [what to focus on] the user Review this branch's changes for security problems: injection, authentication and authorization, secrets, unsafe deserialization, path traversal, SSRF, cryptography misuse, risky dependency changes, data exposure.
/simplify [path or what to look at] the user Look at the changed code for reuse of existing helpers, simpler forms, wasted work and the wrong level of abstraction, then apply the improvements the user accepts. Quality only, not bug hunting.
/review [low|medium|high|max] [--fix] [--comment] [PR number|branch|path] the user The code-review command under its short name.
/ultrareview [PR number|branch] the user The deepest code review, done locally: the code-review skill at level max.
/bisect [bad commit] [good commit] [test command] the user Find the commit that broke something with git bisect, driven by a test command when there is one.
/batch the user Split a large change into independent units and have background agents do them in parallel, each in a git worktree and branch of its own, then report each unit's state.
/autofix-pr [PR number] [what to focus on] the user Watch this branch's pull request and fix what fails: its checks and the review comments, committing and pushing, until they pass or ten rounds went by.
/git-history the user or the model Answer questions about the history of the code: why a line is there, who changed a function and when, when something broke, what changed since a version. Use when the user asks why, when or by whom code changed.
/install-github-app [the runner's label] [the model's path on it] the user Write a GitHub Actions workflow that runs OKLM on a self-hosted runner when someone mentions @oklm in a pull request or issue comment.
/run [what to look at] the user Build, launch and drive the project's application to see a change working: a server answering, a command's output, a page served.
/verify [the change, or what it should do] the user Confirm a change works by running the application and watching the result, not only the tests.
/run-skill-generator [anything to know about running it] the user Write a skill for this project that teaches /run and /verify how to build, launch and drive its application from a clean checkout.
/debug the user Find the cause of a bug methodically: reproduce it, narrow it down, prove the cause, fix it, and show the fix works.
/deep-research the user Research a question across the project's code, documents and history with several sub-agents, and write a cited report.
/docker the user or the model Run, observe and clean up containers with Docker or Podman for this project: compose stacks, one-off containers, their logs, health and exec. Use when the task involves docker, podman, containers or compose.
/dev-server the user or the model Start, watch and reuse the project's development server or watcher (npm run dev, vite, uvicorn --reload, rails s, cargo watch, air...) and catch its compile errors. Use when a task needs the app running locally or a watcher's errors.
/database the user or the model Run a local database or service for development and tests (PostgreSQL, MySQL, Redis, MongoDB, LocalStack), check it is ready, and query it safely. Use when the task needs a database or such a service running.
/playwright the user or the model Open the local application in a headless browser with Playwright: screenshots, the page's text, console errors, and end-to-end test runs. Use when a task needs to see or test a web page.
/kubernetes the user or the model Work with a local Kubernetes cluster (kind, minikube, k3d): create it, deploy, wait for rollouts, read logs and events, port-forward, clean up. Use when the task involves kubectl, helm, kind, minikube, tilt or skaffold.
/android the user or the model Build, install, run and watch an Android app on an emulator: start it headless, wait for boot, install, launch, tap and type, logcat, screenshots, the UI tree. Use when the task involves Android, adb, an emulator or an APK.
/vms the user or the model Run and control virtual machines for testing (QEMU, Vagrant, Multipass, Lima): start headless, wait for SSH, run commands in them, watch the console, stop them. Use when the task needs a VM.
/cloud the user or the model Look at cloud and infrastructure state with aws, gcloud, az, terraform and pulumi, read-only unless the user asks for a change. Use when the task involves a cloud account, terraform or pulumi.
/tunnels the user or the model Expose a local port to the internet or reach a remote one (ngrok, cloudflared, localtunnel, ssh -L and -R), only when the user wants it. Use when the task needs a public URL or a port forward.
/update-config [the change wanted] the user or the model Change OKLM's settings for the user: permission rules, environment variables, hooks, the model, output style, any key of settings.json. Use when the user asks to allow or deny a tool or command, to run something whenever an event happens, or to set a setting.
/fewer-permission-prompts the user Propose allow rules for the read-only commands the user keeps being asked about, from this project's audit logs.

Left out

These serve an account, billing or a cloud service of a hosted agent, or were removed upstream; typed, they say so.

/login, /logout, /upgrade, /usage-credits, /rate-limit-options, /privacy-settings, /passes, /stickers, /radio, /mobile, /ios, /android, /install-slack-app, /web-setup, /remote-env, /setup-bedrock, /setup-vertex, /design, /design-login, /design-sync, /slides, /artifacts, /artifact-capabilities, /artifact-diagramming, /claude-api, /dataviz, /teleport, /remote-control, /desktop, /pr-comments, /ultraplan.

Not built yet

Command Why
/keybindings the key bindings are fixed (no file to edit yet)
/terminal-setup nothing to set up for the line editor yet
/voice speech to text exists (oklm transcribe), nothing records yet
/subtask background sub-agents exist (the Agent tool), not as a command
/fast a second, faster model is not offered yet
/advisor a second model is not offered yet
/workflows with workflows (not built)
/workflow-authoring with workflows (not built)

MCP servers' prompts are commands too, /mcp__SERVER__PROMPT (see /mcp); project and user skills add their own.