Non-interactive mode

A coding agent runs to11 in a shell with nobody at the keyboard. In that shell a question would wait forever, so the CLI does not ask it: it prints what it needs as one JSON object, and the agent either finishes the work with a command from that object or hands one sentence to a person.

When the CLI does not prompt

A run is non-interactive when any of these holds:

Condition	Notes
stdin is not a terminal	there is nobody to type an answer
stderr is not a terminal	questions are written to stderr. stdout carries only the result, so `to11 login \| tee log` still asks
`--non-interactive`	on `to11`, `to11 link`, `to11 login` and `to11 skill import`
`TO11_NON_INTERACTIVE` is set	any value except empty, `0`, `false`, `no` and `off`
an agent is detected	one of the variables below is set

--interactive overrides the flag, the variable and agent detection — for a person working inside an agent’s shell. It never overrides a missing terminal. --interactive with --non-interactive is a usage error. The commands that can prompt are a bare to11, to11 link, to11 login, to11 skill import, and to11 skill sync, which asks one question: whether to remove a skill you picked that is no longer published. to11 ui is a full-screen program and exits 1 without a terminal instead.

Agent markers

An agent is detected from the variables a coding agent sets for the processes it runs. Any non-empty value counts, except where a value is given.

Variable	Set by
`AI_AGENT`	the generic marker
`CLAUDECODE`, `CLAUDE_CODE`	Claude Code
`CURSOR_AGENT`	Cursor
`CURSOR_EXTENSION_HOST_ROLE=agent-exec`	Cursor, this value only
`GEMINI_CLI`	Gemini CLI
`CODEX_SANDBOX`, `CODEX_CI`, `CODEX_THREAD_ID`	Codex
`AUGMENT_AGENT`	Augment
`OPENCODE_CLIENT`	OpenCode

Variables that a person’s own shell also carries are not markers, because they would make a person at a terminal non-interactive: CURSOR_TRACE_ID, set in every Cursor terminal; REPL_ID, set in every Replit shell; COPILOT_GITHUB_TOKEN, COPILOT_MODEL and COPILOT_ALLOW_ALL; and ANTIGRAVITY_AGENT. A flag always wins over detection. A marker changes whether the CLI asks, never what --project, --organization or any other flag means.

What changes

Nothing is asked. Each value to11 link takes is read from its flag, its variable, the location’s file and what can be derived, exactly as at a terminal. Then:

a value that is still missing is the action_required answer below, and nothing is written;
Change targets? is not asked, and the targets are what was detected or given;
Which skills? is not asked, and the file’s per-skill choices are left as they are — unless --skill names the skills to receive, which works the same with or without a terminal: to11 link --project acme/web --skill code-review --skill incident-triage, or --skill '*' for all. A name the project does not have fails the link, and nothing is written;
the plugin offer is skipped, unless --yes was given;
with nobody signed in, nothing is attempted — the answer is not_signed_in;
after the link is written, skill folders already in the targets that to11 did not deliver are not adopted, uploaded or asked about: the answer is unmanaged_skills, and the sync does not run.

The `action_required` object

The command prints one JSON object on stdout, exits 1, and writes nothing. Header lines such as to11 CLI <version> stay on stderr, so stdout parses as exactly one object.

{
  "status": "action_required",
  "reason": "missing_project",
  "message": "Name the project this directory receives skills from.",
  "choices": [
    {
      "name": "acme/web",
      "id": "proj_web",
      "matched": "directory name"
    },
    {
      "name": "acme/platform",
      "id": "proj_platform"
    }
  ],
  "next": [
    {
      "command": "to11 link --project acme/web --yes",
      "when": "Link this repository, then run `to11` with no flags"
    },
    {
      "command": "to11 link --project acme/platform --yes",
      "when": "Link this repository, then run `to11` with no flags"
    }
  ],
  "hint": "Run one of the commands in next[] to complete without prompting."
}

Field	Type	Holds
`status`	string	always `action_required`
`reason`	string	what is missing — one of the reasons
`message`	string	one sentence a person can read
`choices`	array of objects	the values the question would have listed. Absent when there is nothing to choose from
`next`	array of objects	commands that finish the work without prompting. Present only when such a command exists
`ask`	object	what a person has to do first. Present only when no command can finish the work
`hint`	string	one of two fixed sentences, chosen by which of `next` and `ask` is present

Fields come in this order and are absent rather than null. The objects in choices, next and ask are flat: each field is a string. This is the one place the JSON contract nests, and it stays additive only. Exactly one of next and ask is present. next means the caller can finish on its own; ask means it has to stop and hand the sentence to a person.

Object	Field	Holds
`choices[]`	`name`	the organization slug, or the project as `workspace/project`
	`id`	its id
	`matched`	`directory name` when the project’s name matches the repository’s folder. Otherwise absent
`next[]`	`command`	a complete command line
	`when`	what it does: Link this repository, then run `to11` with no flags — or Link your home, then run `to11` with no flags. A `--global` step from inside a repository says to run `to11 skill sync` instead
`ask`	`message`	the sentence to relay to a person
	`url`	where the person does it — the dashboard. Absent for `not_signed_in`

The two hints

Present	`hint`
`next`	`Run one of the commands in next[] to complete without prompting.`
`ask`	`No command completes this without a person. Relay ask.message to one, then run the original command again.`

Both are fixed, so a script can match either string.

Reasons

`reason`	Carries	When
`missing_organization`	`choices`, `next`	you belong to several organizations and nothing names one. `next` has a step for every project, in every organization
`missing_project`	`choices`, `next`	nothing names exactly one project. A single directory-name match counts only with `--yes`
`not_signed_in`	`ask`	nobody is signed in and `TO11_API_KEY` is not set
`no_organization`	`ask`	you are not a member of any organization
`no_projects`	`ask`	the organization has no project you can read
`unmanaged_skills`	`choices`, `next`	the location’s targets hold skill folders to11 did not deliver, and no flag says what to do with them. From `to11 link` and a bare `to11` after the link is written, and from `to11 skill import`
`not_linked`	`next`	`to11 skill import` ran where no location it lists is linked. `next` is `to11 link`
`ambiguous_skill`	`choices`, `next`	a `to11 skill import` flag named a slug waiting in both the repository and your home. Each choice is one folder, with its `location`; each `next` names by path a folder the flag can act on, and a choice it cannot carries a `reason`. When the flag fits neither folder, the command is refused as an error instead

A value you cannot read is the same question. When --organization, --project or their variables name something you cannot read, the reason is the matching missing_ one, message starts with a sentence naming the value — acme has no project acme/nope that you can read. — and choices holds only what you can read. Signing in is always an ask, never a next. Non-interactive to11 login is itself action_required, so a next naming it would point an agent at the command that printed the object it was trying to resolve:

{
  "status": "action_required",
  "reason": "not_signed_in",
  "message": "Nobody is signed in on this machine.",
  "ask": {
    "message": "Run to11 login in a terminal and sign in, then run this command again."
  },
  "hint": "No command completes this without a person. Relay ask.message to one, then run the original command again."
}

`unmanaged_skills`

Its choices are skill folders, each with the location it is in, and its next steps are to11 skill import lines — with --global for your home’s folders, and a path in place of a slug waiting in both locations. A folder in a location that is not linked is a choice with a reason, and that location’s to11 link is a step. The link file is already written when a link stops here; nothing else is.

{
  "status": "action_required",
  "reason": "unmanaged_skills",
  "message": "3 skills in ./.claude/skills were not delivered by to11. Say what to do with each, or skip them.",
  "choices": [
    {
      "name": "commit-style",
      "id": ".claude/skills/commit-style",
      "matched": "acme/web commit-style v4",
      "location": "repo"
    },
    {
      "name": "review-checklist",
      "id": ".claude/skills/review-checklist",
      "matched": "differs from acme/web review-checklist v2",
      "location": "repo"
    },
    {
      "name": "my-notes",
      "id": ".claude/skills/my-notes",
      "location": "repo"
    }
  ],
  "next": [
    {
      "command": "to11 skill import --adopt commit-style --yes",
      "when": "Record it as the published v4; nothing is uploaded"
    },
    {
      "command": "to11 skill import --keep review-checklist --yes",
      "when": "Keep the local copy and never deliver the published one here"
    },
    {
      "command": "to11 skill import --upload review-checklist --yes",
      "when": "Store it in acme/web as v3; whoever follows `latest` receives it on their next sync"
    },
    {
      "command": "to11 skill import --take review-checklist --yes",
      "when": "Replace the folder with the published v2; the files in it now are deleted"
    },
    {
      "command": "to11 skill import --upload my-notes --yes",
      "when": "Store it in acme/web as v1; whoever follows `latest` receives it on their next sync"
    },
    {
      "command": "to11 skill import --skip --yes",
      "when": "Leave every folder as it is; the sync reports them as conflicts"
    }
  ],
  "hint": "Run one of the commands in next[] to complete without prompting."
}

Field	Holds
`choices[].name`	the folder’s name, which is the skill’s slug
`choices[].id`	the folder, relative to the location’s root
`choices[].matched`	the published skill and version the folder is, or `differs from` the one it differs from. Absent for a folder the project does not have
`choices[].reason`	why the folder has no step in `next`, and what to do about it: `needs a description in SKILL.md; add a \`description:` line to the front matter of SKILL.md, then run `to11 skill import` again`. Absent otherwise

next has one step per folder per outcome it can have — adopt for an exact match; keep, upload and take for one that differs; upload for one the project does not have, when a copy of it has a description in its SKILL.md (otherwise the choice carries a reason) — and a last step, --skip, that leaves every folder as it is. Each step acts on the folder it names and leaves the others, so any one of them exits 0. A keep step promises the published skill is never delivered here only when no other target holds a copy of it that is, or can be adopted as, to11’s; otherwise its when says the other targets keep receiving it. Run to11 afterwards to sync: in a linked location it does not look at the folders again. A step carries --project when the location’s file names several projects, and --global when the location is your home and the command runs inside a repository.

The `next` commands

A next[] command finishes without prompting and exits 0. Each carries every value already settled — --organization when you belong to several, --project — and the inputs that shape the file: any --target, --prefix, --skill and --no-plugin you gave, --yes, and --global when the location is your home and the command runs inside a repository. A next[] command never carries a credential. The CLI takes no token flag, so none can reach a suggested command line. After a link, to11 with no flags syncs the location — the when field says so. A home linked from inside a repository is synced with to11 skill sync, because to11 there acts on the repository. After a to11 skill import step, so does to11, or to11 skill sync for that home.

`--yes`

--yes answers confirmations, never data questions. It accepts a single directory-name match, the detected targets and the plugin offer. It does not choose an organization or a project from several: with any of those open, the answer is still action_required.

Next steps

Commands

What to11 link decides, in which order, and where it writes.

Output formats

Plain versus JSON, and what the JSON contract promises.

SDKs

CLI

Gateway

Non-interactive mode

When the CLI does not prompt

Agent markers

What changes

The `action_required` object

The two hints

Reasons

`unmanaged_skills`

The `next` commands

`--yes`

Next steps

Commands

Output formats

​When the CLI does not prompt

​Agent markers

​What changes

​The action_required object

​The two hints

​Reasons

​unmanaged_skills

​The next commands

​--yes

​Next steps

Commands

Output formats

When the CLI does not prompt

Agent markers

What changes

The `action_required` object

The two hints

Reasons

`unmanaged_skills`

The `next` commands

`--yes`

Next steps