In-REPL slash commands
Every built-in slash command in the interactive REPL — session, model roles, memory, history, and interface — plus how the / typeahead merges built-in, CLI, and custom commands.
Inside the interactive REPL (oxagen with no arguments), type / to open a typeahead menu of every command you can run without leaving the session. It merges three tiers into one list:
| Tier | Where it comes from | Marked |
|---|---|---|
| Built-in | REPL-native commands the session interprets directly (/model, /diff, /remember, …). | Lock glyph 🔒 |
| CLI | Every oxagen subcommand, surfaced so you can run it without leaving the REPL. | Lock glyph 🔒 |
| Custom | Your own prompt templates from .oxagen/commands/*.md. | No glyph |
Names are deduped with precedence built-in › CLI › custom, so a REPL-native command always wins over a same-named CLI or custom one.
This page is the reference for the built-in tier. For the CLI tier see the command reference; to write your own, see custom commands.
/name is one of three ways to reach a setting. Most turn-configuration commands also have a --flag for one-shot and headless runs, and a settings.json key for a persistent default. This page documents the interactive form; the flag and key are cross-linked where they exist.
Session
| Command | Does | Example |
|---|---|---|
/help | List the slash commands and keybindings. | /help |
/login | Show the current session, or open the browser sign-in panel if you're signed out. | /login |
/logout | Clear the stored session — token, org, workspace. | /logout |
Project
| Command | Does | Example |
|---|---|---|
/init | Scaffold .oxagen/ and build the local code graph for this repo. | /init |
/config [doctor] | Browse and edit tiered config (repo ▸ workspace ▸ user ▸ org); doctor scans it for problems. | /config doctor |
Turn configuration
These change how the next turn runs. Model-role commands are covered in depth in Per-role models; budget in Per-turn budget; the pipeline and modes in The agent engine.
| Command | Argument | Does |
|---|---|---|
/model | [slug] | Show or set the worker model (alias for /worker-model). |
/worker-model | [slug] | Show or set the worker (executor) model — the one that edits code. |
/judge-model | [slug] | Show or set the judge (completeness advisor) model. |
/triage-model | [slug] | Show or set the triage/coordinator model — the planner and evaluator. |
/coordinator | [remote|local] | Run turns on the remote gateway or an on-device model. |
/effort | [low|medium|high|xhigh|max|default] | Reasoning effort for thinking-capable models. |
/budget | off | <usd> [grace|prompt|enforce] | mode <mode> | status | Per-turn dollar budget. |
/mode | [ask|auto-edit|bypass|readonly] | Tool-permission posture. |
/pipeline | [on|off] | Toggle prompt evaluation, context injection, and judging. |
/verbose | [on|off] | Per-phase timing, token/cost, and tool telemetry. |
Run any of these with no argument to print the current value:
› /worker-model
Current worker model: anthropic/claude-sonnet-5
› /judge-model
Current judge model: (engine default — advisor tier, distinct from worker)
› /mode
Current mode: askSet one by passing an argument:
› /worker-model anthropic/claude-opus-4-8
Worker model set to anthropic/claude-opus-4-8 (saved to .oxagen/settings.local.json).
› /mode auto-edit
Mode: auto-edit — file edits apply automatically; shell commands still ask.
› /effort high
Reasoning effort: highPermission modes
/mode sets what the agent may do without asking:
| Mode | Behavior |
|---|---|
ask | Prompt before each file edit or shell command. (REPL default.) |
auto-edit | Apply file edits automatically; still prompt for shell commands. |
bypass | Run everything without prompting. ⚠ Tool calls run with no confirmation. |
readonly | Read, search, and explain only — file edits and commands disabled. |
Pipeline on/off
/pipeline off runs the bare agent: it skips prompt evaluation, context injection (graph + memory grounding), and the completeness judge — a raw model loop. /pipeline on restores the full pipeline. See The agent engine for what each stage does.
Memory
The workspace's agent memory. Memories are shared across every session in the workspace.
| Command | Argument | Does | Example |
|---|---|---|---|
/remember | <text> | Save a memory; the epistemic class and kind are inferred from the text. | /remember The staging DB is on port 5433 |
/memories | [class|kind] | Browse saved memories, optionally filtered by class (OBSERVATION/RULE/FACT) or a kind. | /memories RULE |
/forget | <id> | Delete a memory by id. | /forget mem_01H… |
History
| Command | Argument | Does | Example |
|---|---|---|---|
/replay | [n|id] | Replay how a past turn was handled — prompt, scores, injected context, model, judge verdict. 1 is the most recent; omit for the latest. | /replay 2 |
/traces | List recent turns you can /replay. | /traces |
Interface
| Command | Argument | Does |
|---|---|---|
/diff | [path] | Review working-tree changes — pick a changed file and view its diff, or open straight to a path. |
/hud | Toggle the running-agents heads-up display (turn, subagents, monitors; time/model/cost each). | |
/panel | Toggle the Agent Team + Task side panel. | |
/mouse | Toggle the CLI's own mouse-wheel scroll. Off by default so your terminal's native copy/paste works. | |
/motion | [full|reduced|off] | Animation level. reduced drops the intro duel and border flash; off disables all motion including the thinking indicator. |
/clear | Clear the conversation history. |
Mouse scroll is off by default on purpose: with it off, the terminal owns text selection so drag-to-select and Cmd/Ctrl-C copy clean text. Keyboard scrolling (PageUp/PageDown, Ctrl-U/Ctrl-D, arrows on an empty input) always works. See Copy, paste, and scroll.
Lifecycle
| Command | Does |
|---|---|
/exit | Quit the REPL. |
/quit | Quit the REPL (alias of /exit). |
Dispatching a fleet with &
Not a slash command, but the other REPL shortcut worth knowing: end a prompt with a trailing & and — just like a shell — it fans out to the fleet as a detached session instead of running inline, so your composer stays free:
› fix the flaky login test &
◇ dispatched fs_01J… to the fleet — `oxagen fleet` to watch it work,
`oxagen fleet send fs_01J… "…"` to follow up.The prompt runs in the background; watch it with oxagen fleet in any terminal. (A prompt starting with / or !, or a bare &, is never treated as a dispatch.)
Related
- Custom slash commands — author your own
/nametemplates. - Per-role models — set the worker, judge, and triage models.
- The agent engine — what
/pipeline,/mode, and scope review actually control. - Command reference — the CLI tier of the slash menu.
Command reference
Every oxagen command and flag — the agentic loop, fleets, the knowledge graph, settings, agents, rules, MCP servers, environments, and secrets.
The agent engine
How one prompt becomes planned, executed, and verified work — the evaluate → enhance → route → execute → judge → revise pipeline, the three model roles, scope-review approvals, and the fast-paths that skip work you don't need.