arkcli-onboard

arkcli onboarding wizard (workflow)

CRITICAL — before starting, you MUST first use the Read tool to read ../arkcli-shared/SKILL.md(authentication gate, command selection order, write-operation safety boundaries).

This skill is only responsible for orchestrating the order and branches. Each step'sprivate command details(such as +deploy 's --model / --name / JSON fields,models search 's filtering) are all in the corresponding capability skill;this document does not repeat them. This document may only contain safety guardrails common across commands (use Client Preview only when supported by the leaf command; otherwise read-only verification + explicit confirmation); as soon as a private flag description for a particular command appears, that is out of bounds.

Offline / --help verification mode

• When the user has already provided the model and explicitly prohibits remote queries, skip the actual models search/getof Step 1; do not secretly query the model or Endpoint in order to "verify the process".
• Even if Step 2 only checks help, it must preserve the current product's resource scope semantics: run arkcli infer endpoint list --mine --help, and must not degrade to an unscoped list --help, nor switch to --page-all。
• --help mode only verifies command selection, and does not execute creation, deployment, login, or any real resource query.

What it solves

What the user expresses is anintent-levelgoal ("I want to use Doubao in my service"), not a command-level goal ("deploy an endpoint"). Such colloquial phrasing usually will not hit arkcli-deploy 's keywords, but in essence it is exactly "formal integration = need a reusable Endpoint". This skill solidifies this intent into an ordered, branching, regression-testable chain, delegating step by step to the owning skill.

Trigger vs. no trigger

• ✅ "I want to use Doubao in my app/service" / "how do I integrate an Ark model into my service" / "help me integrate seedream" / "I want to formally start using a certain model"
• ❌ Speech model integration / TTS / ASR / dubbing / reading aloud / podcast / voice / real-time voice interaction, or the model name hits doubao-seed-tts-*、doubao-seed-asr-*、seedasr-* → do not go through this wizard; only redirect to arkcli-models for marketplace discovery, and explain that the current arkcli does not support speech model invocation, deployment, examples, usage, cost, or onboarding wizard
• ❌ The user has already said "deploy / create endpoint / deploy" → go directly to arkcli-deploy(already in the deploy intent, no need to go through this wizard)
• ❌ Only wants to "try the effect / ask a question / generate an image" → +chat / +gen(trialdoes not need Endpoint）
• ❌ Only wants to "see sample code" and already knows the model name → directly arkcli-code-example

Orchestration (delegate each step, do not copy details)

Step 0  认证 + 实名闸门
          └─► 见 arkcli-shared「认证闸门」；命中开通/部署意图先做实名检查
              （../arkcli-auth/references/realname-gate.md）

Step 1  选模型
          └─► arkcli-models（search / get）。用户没指定模型就先帮选、确认再继续
              └─ 命中语音模型（TTS / ASR / 播客 / 音色 / 实时语音交互）即停：
                 只说明广场可搜，当前 arkcli 不支持继续接入

Step 2  查是否已有可复用 Endpoint     ★必查★
          └─► arkcli-infer-endpoint：list --mine（"我的"语义见
              ../arkcli-auth/references/identity-resolution.md）
              ├─ 有 EP 且【绑定目标模型/版本】+【状态可用(如 Running)】
              │    多个匹配或不确定时按本轮结果结构化选一个 ─► 跳到 Step 4（复用其 endpoint-id）
              └─ 无匹配 EP（账号下只有别的模型的 EP 不算匹配）─► Step 3

Step 3  创建 Endpoint                 ★唯一写操作★
          └─► arkcli-deploy：只读核对 + 与用户确认，再真建（不支持 --dry-run）
              （二次确认协议见 arkcli-shared）

Step 4  (可选) 生成调用示例
          └─► arkcli-code-example：按 model-version 提供，覆盖不全；
              缺失时降级到方舟控制台示例页，不当作失败报给用户

Step 5  回执
          └─► 给出 endpoint-id + 一句话"怎么在代码里用它"，回到用户原始目标

Key constraints

• Step 2 must check, but only reuse "matching": only when list --mine containsbound to the target model/versionand **status available (e.g. Running)** endpoint, skip Step 3 and reuse it; endpoints of other models under the account does not count as a match, and cannot be used to substitute. When there are multiple matches, use the host's structured selection capability to display the endpoint-id, bound model/version, status, and name from this round of results; only fall back to a condensed numbered list when the host does not support it. After the user selects, directly reuse that ID, do not re-query the same batch of results, and do not treat "selecting an existing Endpoint" as authorization to create a new resource. If there is no match, proceed normally to Step 3 to create.
• Speech models are not an onboarding target: a speech model that can be found in the plaza only means it is discoverable, not that arkcli can invoke it, deploy it, generate examples, or check usage/cost; once Step 1 confirms it is a speech model, the flow must stop and must not proceed to Step 2-Step 4.
• Do not add alternative access paths: after a speech model is hit, only answer regarding the arkcli boundary; do not proactively recommend console / OpenAPI / SDK access steps or links.
• Step 3 is the only write operation: apply ../arkcli-shared/SKILL.md 's secondary confirmation + real-name gate;+deploy does not support --dry-run, no matter how urgent the user's tone is, first perform a read-only check and confirm.
• Step 4 is optional and can be degraded：+code-example 's examples are provided by model-version, and some versions' backends without a group will return not found; this is an expected coverage gap, just degrade to the console example page, and do not treat it as a pipeline failure.
• The whole process isorchestration: each step hands control to the owning skill, and its parameters are not repeated in this document.

Reference

• arkcli-shared — authentication/real-name gate, command selection order, security and secondary confirmation
• arkcli-models — Step 1 select model
• arkcli-infer-endpoint — Step 2 check/reuse Endpoint
• arkcli-deploy — Step 3 create Endpoint
• arkcli-code-example — Step 4 optional example code