CLI reference

This page is the output of wfx help on main, reorganised by verb. It is read from the same string the binary prints, so a command that is not here does not exist.

Verbatim

wfx — run agent workflows over a repository

  wfx workflows                    list workflows
  wfx workflows show <name>        every step's harness: skills, tools, team, budget, gates
  wfx workflow proposals [--project p] [--workflow w]  pending workflow changes (git branches/PRs)
  wfx workflow approve <id>        merge a proposal; the project's checkout pulls
  wfx workflow reject <id> --reason "why"  close its PR and delete its branch
  wfx workflow drift [--project p] workflows changed on disk but never committed
  wfx workflow drift --propose <w> [--project p]  open a proposal from that on-disk change
  wfx apply <file.yaml> [--project p]  validate and install a workflow; a git-backed project gets a proposal
  wfx apply --from-run <id>        install the workflow a run authored (validated first)
  wfx validate <file.yaml>         validate only; writes nothing
  wfx run <workflow> -i k=v [-f] [-v]  start a run (-f follows it; -v every tool call)
  wfx runs [--project p] [--workflow w]  recent runs, newest first
  wfx login --url <host>           sign in to a host (prints a code; no browser needed here)
  wfx login --url <host> --user N [--provider ldap-name]  sign in with a password (WFX_PASSWORD, prompt or stdin)
  wfx login --approve <code>       approve a sign-in code shown elsewhere (e.g. the web UI)
  wfx logout [--url <host>]        revoke this host's token and forget it locally
  wfx user ls | add NAME --role R | disable NAME | passwd [NAME] | role NAME R   users (admin)
  wfx member add|rm|ls USER --project P | --workflow W [--role R]   scoped roles
  wfx role ls                      the four roles
  wfx ui [--url <host>] [--print] open the web UI signed in (one-time link, 60s)
  wfx profile create <name> --url U [--via ssh://user@host [--key k] [--accept-host-key]]
                                   register a server you can already reach (kubectl-style)
  wfx profile create <name> --install local|ssh://user@host|k8s [--compose|--image] [--version v] [--dry-run]
                                   provision the server (install.sh, docker compose, GHCR image or helm),
                                   register it, log in
  wfx profile ls | use|switch <name> | current | rm <name>   * marks the current one
  wfx context add|use|ls|rm …      the same profiles under their older name, kept as an alias
  wfx status                       current context; for a via context, its tunnel
  wfx tunnel stop [context]        close a context's tunnel now (else 15 min idle)
  wfx team add <name> <pubkey> | remove <name> | list   on the server: who may tunnel in
  wfx publish <file.yaml> --version v  push a workflow and its skills to the host
  wfx publish <file.yaml> --version v --to <git remote>  publish to a git repo instead (no login)
  wfx pull <digest>                materialise a published bundle on the host
  wfx install --skills             install every agent skill into ~/.claude and ~/.agents
  wfx worker install --to local|ssh://user@host|k8s [--image] [--version v] [--dry-run]
                                   put a runner on a machine; then "wfx workers" for its join line
  wfx skill add <name|git-url|path> [--ref r]  agent skills onto the current profile's server; also rm, ls
  wfx mcp add <name> --set command="npx -y pkg" | --set url=U [--env NAME]…  an MCP server; also rm, ls
                                   --env names an env-store key; its value is injected at run time, never stored here
  wfx tool add <name>[@v] [--on worker]   install a tool through the server's carry; also rm, ls
  wfx config put <abs-path> <file> [--on worker] [--to ctx]  place a config file; also get, ls, rm
  wfx config env set K=V | unset K | ls   non-secret env defaults (names only are ever shown)
  wfx config log | rollback <n> | status  carry history, roll back, is carry on
  wfx sync --from <ctx|local:DIR> --to <ctx|local:DIR> [--only files,env] [--dry-run]
                                   move config files and env defaults between places
  wfx version [--json]             which binary this is, and where it came from
  wfx install --list               what this platform ships
  wfx env [--project p]            the platform's env store (values are never shown)
  wfx env set NAME [--plain]       read a value without echoing it, store it encrypted
  wfx state list --workflow        what this workflow remembers between runs
  wfx state set --workflow k v     remember it (also --step, --project, --global)
  wfx state get --step k           one value; empty when never written
  wfx memory ls --workflow w       what previous runs of a workflow left behind (on by default)
  wfx memory show --workflow w     the section the next run's agent steps will read, then every entry
  wfx memory forget --workflow w   delete a workflow's memory (or: wfx memory forget <id>)
  wfx memory add --workflow w "t"  write a lesson the next run's agent steps will read
  wfx memory edit <id> "text"      rewrite one entry's text
  wfx expose hook|mcp <wf>         a token-protected hook URL / MCP endpoint for one workflow (--rotate, --revoke)
  wfx expose mcp --project p       one MCP token for every opted-in workflow of a project
  wfx prices                       what each model costs per 1M tokens (approx until set)
  wfx prices set MODEL IN OUT      set it, e.g. claude-sonnet 3 15; MODEL * is the fallback
  wfx prices rm MODEL              drop it; a built-in family returns to its default
  wfx templates [name]             starting points to copy — the shape, minus your skills
  wfx new <template> --as <name>   copy one into a workflow of your own
  wfx adapters                     agents wfnexus drives over ACP (devin, opencode, codex) and whether each is installed
  wfx models <provider> [--free]   the models an acp provider's agent offers, asked live
  wfx workers                      machines in the pool, and the line that adds another
  wfx workers rm <id>              forget a machine
  wfx workers rotate               new join token; machines already joined keep working
  wfx show <run-id>                a run, step by step
  wfx logs <run-id> [-f] [-v]      the activity log (-v: full transcript)
  wfx approve <run-id> [--as who] approve the step waiting on a human
  wfx reject <run-id> -m "why"     reject it
  wfx answer <run-id> -m "text"    answer an agent's question
  wfx send [--channel c] [--file f | --url u] [--target t] text
                                   send on a channel (the server default if omitted); same path as a send: step
  wfx messages [--channel c] [--run id]  messages received on channels, newest first, with what each did
  wfx channels                     channels, which is the default, inbound or not, allow-list size, last receive error
  wfx channels receive <name>      poll an inbound channel now instead of waiting for its interval
  wfx sub ls|show|add|update|pause|resume|rm|validate   subscriptions: which chat messages start which workflow (ADR 0026)
  wfx dryrun <workflow> [-i k=v]   would it run here? no model, no repo, no writes
  wfx eval <workflow> --corpus f.yaml --providers a,b [--json]  the portability matrix: same workflow, N backends, pass/tokens/$
  wfx judge -q q.yaml --items f.jsonl  calibrated typed questions per item (JEV); bands no|uncertain|yes
  wfx projects                     every project: workflows and run activity
  wfx project add <repo> [--as n] [--category c]  add a repo that has .wfx/workflows/
  wfx project new <name> [--dir path] [--category c]  start an empty project
  wfx project set <name> --category c  what kind of work it does (wfx project categories)
  wfx project rm <name>            forget one (the clone stays on disk)
  wfx sources [forget <name>]      where workflows are loaded from
  wfx doctor                       what is wired: default model, providers, classifiers, skills
  wfx retry <run-id> [--step id]   re-run from a step
  wfx cancel <run-id>
  wfx registry [skills|tools|providers|classifiers|mcp]
  wfx skills import <git-url> [--ref r] [--path skills/] [--name n]  import a git repo of skills; r is a branch, tag or commit (--branch = --ref)
  wfx module ls | install <git-url> [--ref r] | sync <source> | config <module>   modules: task + module.yaml from the registry (ADR 0028)
  wfx skills sources               skill sources: url, ref, commit, skill count and names
  wfx skills ref <source> <ref>    switch a source to another branch, tag or commit (alias: branch)
  wfx skills refresh <source>      pull a branch source again; a tag or commit is pinned
  wfx skills remove <source>       unload its skills and delete the checkout
  wfx skills show <source>         a source's skills

The API is --url, then $WFX_API, then the current context, then
http://127.0.0.1:8090. Contexts live in ~/.config/wfx/contexts.json (0600).
Add --json to any listing for machine-readable output.

By verb

wfx adapters

Command	What it does
wfx adapters	agents wfnexus drives over ACP (devin, opencode, codex) and whether each is installed

wfx answer

Command	What it does
wfx answer <run-id> -m "text"	answer an agent’s question

wfx apply

Command	What it does
wfx apply <file.yaml> [--project p]	validate and install a workflow; a git-backed project gets a proposal
wfx apply --from-run <id>	install the workflow a run authored (validated first)

wfx channels

Command	What it does
wfx channels	channels, which is the default, inbound or not, allow-list size, last receive error
wfx channels receive <name>	poll an inbound channel now instead of waiting for its interval

wfx config

Command	What it does
wfx config put <abs-path> <file> [--on worker] [--to ctx]	place a config file; also get, ls, rm
wfx config env set K=V | unset K | ls	non-secret env defaults (names only are ever shown)
wfx config log | rollback <n> | status	carry history, roll back, is carry on

wfx context

Command	What it does
wfx context add|use|ls|rm …	the same profiles under their older name, kept as an alias

wfx doctor

Command	What it does
wfx doctor	what is wired: default model, providers, classifiers, skills

wfx dryrun

Command	What it does
wfx dryrun <workflow> [-i k=v]	would it run here? no model, no repo, no writes

wfx env

Command	What it does
wfx env [--project p]	the platform’s env store (values are never shown)
wfx env set NAME [--plain]	read a value without echoing it, store it encrypted

wfx eval

Command	What it does
wfx eval <workflow> --corpus f.yaml --providers a,b [--json]	the portability matrix: same workflow, N backends, pass/tokens/$

wfx expose

Command	What it does
wfx expose hook|mcp <wf>	a token-protected hook URL / MCP endpoint for one workflow (–rotate, –revoke)
wfx expose mcp --project p	one MCP token for every opted-in workflow of a project

wfx install

Command	What it does
wfx install --skills	install every agent skill into ~/.claude and ~/.agents
wfx install --list	what this platform ships

wfx judge

Command	What it does
wfx judge -q q.yaml --items f.jsonl	calibrated typed questions per item (JEV); bands no|uncertain|yes

wfx login

Command	What it does
wfx login --url <host>	sign in to a host (prints a code; no browser needed here)
wfx login --url <host> --user N [--provider ldap-name]	sign in with a password (WFX_PASSWORD, prompt or stdin)
wfx login --approve <code>	approve a sign-in code shown elsewhere (e.g. the web UI)

wfx logout

Command	What it does
wfx logout [--url <host>]	revoke this host’s token and forget it locally

wfx logs

Command	What it does
wfx logs <run-id> [-f] [-v]	the activity log (-v: full transcript)

wfx mcp

Command	What it does
wfx mcp add <name> --set command="npx -y pkg" | --set url=U [--env NAME]…	an MCP server; also rm, ls

wfx member

Command	What it does
wfx member add|rm|ls USER --project P | --workflow W [--role R]	scoped roles

wfx memory

Command	What it does
wfx memory ls --workflow w	what previous runs of a workflow left behind (on by default)
wfx memory show --workflow w	the section the next run’s agent steps will read, then every entry
wfx memory forget --workflow w	delete a workflow’s memory (or: wfx memory forget <id>)
wfx memory add --workflow w "t"	write a lesson the next run’s agent steps will read
wfx memory edit <id> "text"	rewrite one entry’s text

wfx messages

Command	What it does
wfx messages [--channel c] [--run id]	messages received on channels, newest first, with what each did

wfx models

Command	What it does
wfx models <provider> [--free]	the models an acp provider’s agent offers, asked live

wfx module

Command	What it does
wfx module ls | install <git-url> [--ref r] | sync <source> | config <module>	modules: task + module.yaml from the registry (ADR 0028)

wfx new

Command	What it does
wfx new <template> --as <name>	copy one into a workflow of your own

wfx prices

Command	What it does
wfx prices	what each model costs per 1M tokens (approx until set)
wfx prices set MODEL IN OUT	set it, e.g. claude-sonnet 3 15; MODEL * is the fallback
wfx prices rm MODEL	drop it; a built-in family returns to its default

wfx profile

Command	What it does
wfx profile ls | use|switch <name> | current | rm <name>	* marks the current one

wfx project

Command	What it does
wfx project add <repo> [--as n] [--category c]	add a repo that has .wfx/workflows/
wfx project new <name> [--dir path] [--category c]	start an empty project
wfx project set <name> --category c	what kind of work it does (wfx project categories)
wfx project rm <name>	forget one (the clone stays on disk)

wfx projects

Command	What it does
wfx projects	every project: workflows and run activity

wfx publish

Command	What it does
wfx publish <file.yaml> --version v	push a workflow and its skills to the host
wfx publish <file.yaml> --version v --to <git remote>	publish to a git repo instead (no login)

wfx pull

Command	What it does
wfx pull <digest>	materialise a published bundle on the host

wfx reject

Command	What it does
wfx reject <run-id> -m "why"	reject it

wfx retry

Command	What it does
wfx retry <run-id> [--step id]	re-run from a step

wfx role

Command	What it does
wfx role ls	the four roles

wfx run

Command	What it does
wfx run <workflow> -i k=v [-f] [-v]	start a run (-f follows it; -v every tool call)

wfx runs

Command	What it does
wfx runs [--project p] [--workflow w]	recent runs, newest first

wfx show

Command	What it does
wfx show <run-id>	a run, step by step

wfx skill

Command	What it does
wfx skill add <name|git-url|path> [--ref r]	agent skills onto the current profile’s server; also rm, ls

wfx skills

Command	What it does
wfx skills import <git-url> [--ref r] [--path skills/] [--name n]	import a git repo of skills; r is a branch, tag or commit (–branch = –ref)
wfx skills sources	skill sources: url, ref, commit, skill count and names
wfx skills ref <source> <ref>	switch a source to another branch, tag or commit (alias: branch)
wfx skills refresh <source>	pull a branch source again; a tag or commit is pinned
wfx skills remove <source>	unload its skills and delete the checkout
wfx skills show <source>	a source’s skills

wfx sources

Command	What it does
wfx sources [forget <name>]	where workflows are loaded from

wfx state

Command	What it does
wfx state list --workflow	what this workflow remembers between runs
wfx state set --workflow k v	remember it (also –step, –project, –global)
wfx state get --step k	one value; empty when never written

wfx status

Command	What it does
wfx status	current context; for a via context, its tunnel

wfx sub

Command	What it does
wfx sub ls|show|add|update|pause|resume|rm|validate	subscriptions: which chat messages start which workflow (ADR 0026)

wfx team

Command	What it does
wfx team add <name> <pubkey> | remove <name> | list	on the server: who may tunnel in

wfx templates

Command	What it does
wfx templates [name]	starting points to copy — the shape, minus your skills

wfx tool

Command	What it does
wfx tool add <name>[@v] [--on worker]	install a tool through the server’s carry; also rm, ls

wfx tunnel

Command	What it does
wfx tunnel stop [context]	close a context’s tunnel now (else 15 min idle)

wfx user

Command	What it does
wfx user ls | add NAME --role R | disable NAME | passwd [NAME] | role NAME R	users (admin)

wfx validate

Command	What it does
wfx validate <file.yaml>	validate only; writes nothing

wfx version

Command	What it does
wfx version [--json]	which binary this is, and where it came from

wfx workers

Command	What it does
wfx workers	machines in the pool, and the line that adds another
wfx workers rm <id>	forget a machine
wfx workers rotate	new join token; machines already joined keep working

wfx workflow

Command	What it does
wfx workflow proposals [--project p] [--workflow w]	pending workflow changes (git branches/PRs)
wfx workflow approve <id>	merge a proposal; the project’s checkout pulls
wfx workflow reject <id> --reason "why"	close its PR and delete its branch
wfx workflow drift --propose <w> [--project p]	open a proposal from that on-disk change

wfx workflows

Command	What it does
wfx workflows	list workflows
wfx workflows show <name>	every step’s harness: skills, tools, team, budget, gates