shibaox with no command opens the Dashboard. Every command takes --json (one JSON object per line), --remote <url> (a daemon on another machine, see Remote daemon) and --help.

Command
shibaox run <workflow> [--org <dir>] [--project <dir>] [--input <text>] [--issue <n|url>] [--model <ref>] [--adapter mock|direct|claude-code] [--workspace worktree|inplace] [--budget <usd>] [--graph auto|off] [--setup auto|off|<command>] [--detach]	submits a run and follows it (--detach returns at once). --issue makes a GitHub issue the request and reports back on it (GitHub loop). A worktree run installs the project's dependencies first (--setup: shibaox.yaml setup or the lockfile's install by default, off, or a command). The org is ./org when it exists, else the daemon's default org; the project is the current directory. Exit 0 when the run completes, 2 otherwise.
shibaox runs	every run the daemon knows
shibaox follow <runId>	the run view (or plain lines) of a run
shibaox replay <runId> [--db <path>]	the event log and the derived state
shibaox files <runId> [path] [--out <file>]	the files a run created or changed (status, size, lines); with a path, that file to stdout (text) or saved with --out (binaries too); never a protected file (.env, .git, the project's protected: globs)
shibaox audit <runId> [--format md|json] [--out <file>]	everything that happened in a run: request, nodes with attempts and tool calls (with durations), gates with evidence, decisions, who approved what through which channel, git results, cost per node
shibaox runs prune --before <30d|12h|45m|ISO>	removes finished runs older than that, with their events and tool calls; a run whose task is still out stays; worktrees and shibaox/<runId> branches are not touched (shibaox worktree rm)
shibaox resume <runId> [--budget <usd>] [--adapter …]	continues a paused or interrupted run
shibaox cancel <runId>	cancels a run
shibaox steer <runId> <note…> [--node <id>]	redirects the running task: it stops and starts again with the note (the node when several tasks run at once); only a task with a live model call can be steered

Command
shibaox inbox	what waits for you: human:<runId>:<node> and approval:<id>
shibaox approve <id> [--note <text>], shibaox deny <id> [--note <text>]	answer one item; the first answer wins

Command
shibaox init [dir] [--stack node|python|php-laravel|go|auto]	scaffolds org/ and vault/; with a stack also shibaox.yaml (setup, tests, lint, typecheck, protected), typecheck and stack review gates, a weekly security-scan workflow + routine, and (node) a frontend role; auto reads the stack off the project's manifest
shibaox tiers [--org <dir>] [--routing on|off], shibaox tiers set <strong|cheap|decision|judge|adapter|budget> <value>	the org's tiers, judge, adapter, budget and Jev routing (none clears judge/adapter; --routing turns Jev routing of chat turns on or off)
shibaox keys list|set <NAME> [value]|unset <NAME>	the key vault (value from stdin when omitted); a value the daemon refuses (HIGGSFIELD_API_KEY without its colon) prints why and exits 1
shibaox mcp list [--org <dir>], shibaox mcp test <id> [--org <dir>]	the catalog's MCP servers (roles, keys), and a start-and-list check of one on the daemon
shibaox mcp add <id> (--url <https> | --command <cmd> [--arg …]) [--key NAME…] [--header K=V…] [--bearer-command <arg>…] [--tool …] [--role …] [--description …] [--timeout <ms>] [--replace] [--org <dir>], shibaox mcp rm <id> [--org <dir>]	writes catalog/<id>.yaml and gives it to roles (refused when the id exists, unless --replace; --bearer-command takes one argument per flag); rm takes it off every role first (MCP and skills)
shibaox skills list [--org <dir>], shibaox skills add <owner/repo[/path] | git URL | ./dir | /dir | ~/dir> [--folder] [--id <id>…] [--path <p>] [--org <dir>], shibaox skills add <id> --builtin [--replace], shibaox skills rm <id> [--detach] [--org <dir>]	the org's skills: install from a repository (shallow clone; github.com/ in front is fine) or a folder (a source starting with ., / or ~, or --folder), existing ids skipped; --builtin copies a skill shipped with Shibaox (higgsfield, higgsfield-app), --replace rewrites its SKILL.md (--replace without --builtin is refused); rm is refused while a role uses the skill unless --detach
shibaox roles list [--org <dir>]	each role's model, MCP servers and skills
shibaox plugins	Higgsfield, GitHub, Telegram, TypeSafe / Jev: each check, and ready, partial or off; Higgsfield shows its mode (mode auto → api) and both modes ([api] active ready) with their checks
shibaox plugins higgsfield-mode <auto|account|api>	how Higgsfield generates (partners.higgsfield.mode in daemon.yaml); the next task uses it
shibaox providers list [--configured], shibaox providers test <provider> [--model <m>]	the catalog and a real test call
shibaox models --org <dir>	how each role of the org resolves to a model

Command
shibaox app [--no-open] [--port <n>]	the browser app: opens it against this daemon (or a remote one); a socket-only daemon gets a loopback bridge that lives while the command runs (The app)

Command
shibaox daemon start [--detach], status, stop [--force]	the daemon
shibaox daemon install, uninstall	the service: launchd (macOS) or systemd --user (Linux)
shibaox serve [--host <addr>] [--port <n>]	the daemon in the foreground, reachable over the network with the token in SHIBAOX_DAEMON_TOKEN (vault or environment)
shibaox remote set <url> [token], show, clear	send every command to a daemon on another machine (~/.shibaox/remote.json, 0600; the token from stdin when omitted)
shibaox routine add <workflow> --on <trigger> --org <dir> --project <path> [--input] [--name] [--description] [--model <ref>] [--approvals inbox|auto|skip] [--label] [--repo] [--branch] [--every <s>] [--mode always|on_change] [--max-daily <usd>] [--adapter] [--budget], update <id> [the same flags except --org; "" clears a field], list, show <id>, run <id>, pause <id>, resume <id>, rm <id>, sync --org <dir>	what the daemon does on its own: --on cron:<expr>, github:issues|prs|checks, url:<https://…>, file:<path>, command:<cmd>, manual; show says when it runs next and how its last run went; see Routines
shibaox schedule add "<cron>" <workflow> [--org] [--project] [--input], list, rm <id>, run <id>	the cron subset of routines, under its older name
shibaox doctor	prerequisites, the daemon, the service, the install, who decides (the home org's decision tier and whether it is usable), Higgsfield (the CLI, the account, the credits; higgsfield api: the key and whether Higgsfield accepts it, and the mode; a higgsfield skill warning when the org's skill predates the API mode), keys, Telegram, the Claude login
shibaox upgrade	updates an installer checkout and restarts the daemon

Command
shibaox worktree list --project <dir>, shibaox worktree rm <runId> --project <dir> [--delete-branch]	run worktrees
shibaox graph build|update|query "<question>" --project <dir>	the graphify knowledge graph