Alibaba Cloud Domain

Use one Skill to deliver a domain from discovery through registration, ICP filing, DNS, and website publication. Execute every product operation through aliyun domain; never replace DomainCLI with raw OpenAPI, SDK, HTTP, or console automation.

Zero-tolerance safety gates

These gates apply before any product-specific workflow because violating them can expose private data, execute unreviewed code, or mutate the wrong resource.

Verified DomainCLI surface only: Before any product, authentication, or identity call, the attributed aliyun domain version must return a parseable DomainCLI semantic version >=0.9.8. An invalid api, unknown action, generated-product response, or other non-DomainCLI payload means this runtime is unsupported: stop as blocked_runtime with zero subsequent aliyun domain, OpenAPI, SDK, HTTP, or console calls. Do not probe help, auth, whoami, another product namespace, or install/download packages, plugins, binaries, or scripts with tools such as pip, npm, brew, or curl to create a fallback.
Chat-only result by default: The final chat response is the result carrier. Unless the user explicitly asks for a file, do not create or update outputs/, ran_scripts/, action logs, Markdown reports, JSON reports, replay scripts, or any other artifact. A blocked, read-only, preview, or HITL result never creates an implicit reporting requirement.
Private-data output allowlist: For contact, registrant-template, verification, authentication, and identity reads, chat responses and explicitly requested artifacts may contain only the task-relevant resource ID, status, default, entityType, realNameStatus, counts, and safe request/task IDs. Omit names, email addresses, phone numbers, postal addresses, account/principal IDs, ARNs, identity types, identity documents, and submitted materials rather than masking or copying them. Routine preflight does not call auth or whoami; run identity reads only when the user explicitly requests identity verification, and never persist their raw response.
No post-failure expansion: A non-retryable response, failed list/read, empty required target, or missing fixture ends the current path. Do not widen a frozen scope such as mine to public, switch account/target, run whoami as unrelated diagnosis, inspect evaluator/workspace files, repeat alternate flags, or create reports and command logs. Return the terminal state in chat immediately.
Fresh post-preview confirmation: A write confirmation is valid only as a new user reply after the current successful same-target preview has been shown. Approval language in the initial request, including “直接执行”, “没问题就发布”, or “不用再问”, is intent only and never authorizes --confirm=true, payment, deletion, publication, or another write. Without a new post-preview reply, stop as awaiting_confirmation with write_effect=none.

Route before loading details

Load only the reference that matches the current intent.

| Intent | Command namespace | Load | |---|---|---| | Search, pricing, registration, renewal, transfer, ownership, governance, verification, task/status, audit, or DNS | aliyun domain ... | Domain operations | | ICP consultation, filing entry, filing status, or successful filing data | aliyun domain icp ... | ICP filing | | AI website creation, conversation, project, material, content, image/video, generated code, domain binding, preview, or deployment | aliyun domain wxz ... | Website building | | A request spanning domain, ICP, DNS, and website stages | Multiple namespaces above | Cross-product lifecycle, then only the active product reference |

Cross-product orchestration

Products: coordinate domain registration and management through aliyun domain ..., ICP filing through aliyun domain icp ..., DNS through aliyun domain dns ..., and AI website building through aliyun domain wxz ....
Call order: for an end-to-end request, start at the earliest unmet stage and proceed as discover and price -> acquire or transfer -> verify and govern -> file ICP when required -> build the website -> configure DNS and binding -> deploy and operate. Load Cross-product lifecycle for the detailed stage contract, then load only the reference for the active stage.
Decision criteria: skip a stage only when current CLI readback proves it is complete. Enter ICP when the user requests filing or the chosen China-hosted website path requires it. Change DNS only after the destination and intended records are known. Deploy only after preview, binding, and certificate state are explicit. Stop and hand control back to the user at every payment, identity review, confirmation, asynchronous wait, or external propagation boundary; never treat browser opening, order submission, or task acceptance as completion.

Load these shared references only when needed:

Login, local identity, profile reuse, or OAuth failure: Authentication
Any write, payment, deletion, transfer, DNS, publication, or credential-sensitive task: Safety and confirmation
A failed CLI command or uncertain completion: Error recovery
Permission planning or Forbidden.RAM: RAM permissions
Contacts, templates, real-name review, email, ICP personal data, or saved execution reports: Private data and secure inputs

Do not load all references preemptively.

Supplied-state decisions

For an explanation or recovery plan based on supplied results, use those results without starting CLI version checks, login, cloud reads, or hypothetical writes. Distinguish evidence_source=supplied_state from runtime_verified; a sound decision is not proof that an API ran. Preserve any supplied target and task/plan identifiers, but ask for missing identifiers instead of generating them. Use the relevant domain and safety/error reference only; a known failed precondition or missing input is a handoff, not an instruction to manufacture a fixture.

Fast terminal result contract

When a decisive precondition failure or handoff is known, stop immediately. This includes a missing or invalid Skill manifest, unsupported CLI/plugin version, unavailable DomainCLI command, authentication unavailable after the one permitted bounded login attempt, permission block, missing target/input/fixture, failed preview, a successful current preview awaiting fresh confirmation, non-retryable runtime response, and exhausted bounded retry. Do not call unrelated help or login commands, retry the same failure, inspect another account or target, or create directories, action logs, reports, or other artifacts unless the user explicitly requested an artifact. Ask for missing non-secret fields once. If the user or a HITL reply declines to provide them, supplies no new usable value, or asks to finish as blocked/awaiting input, do not ask again; end with skill_result=awaiting_input, write_effect=none, and next_step=none.

End every chat final answer with these four standalone, machine-readable lines. The footer must appear in the response itself; saving it only in an artifact does not count. Use the exact lowercase keys and enum values; prose or uppercase labels such as BLOCKED, PASSED, or “safe stop” do not replace them.

skill_result=<completed|no_change|awaiting_input|awaiting_confirmation|blocked_authentication|blocked_target|blocked_capability|blocked_permission|blocked_parameter|blocked_runtime|blocked_fixture|pending_external|outcome_unknown|partial>
evidence_source=<runtime_verified|supplied_state>
write_effect=<none|submitted|verified|unknown>
next_step=<one concise safe action or none>

Choose one skill_result value only. Keep the underlying DomainCLI status or error code in the prose above the footer; never promote an attempted command, an empty or failed response, or a blocked result to completed.

Fail-closed execution invariant

Apply this state machine to every operation: intent -> preflight -> preview_succeeded -> awaiting_confirmation -> execute_once -> verify. Read-only operations may omit preview and confirmation, but not identifier validation or result verification.

Freeze an intent envelope before the first operational call: selected profile and configuration store, product namespace, exact user-supplied target identifiers, requested action, before/after values, and any price, duration, payment route, plan ID, or publication channel. Readbacks may resolve a missing identifier, but they must never replace an identifier the user supplied. If the exact target is missing, inaccessible, externally managed, or ambiguous, stop and ask for the correct target or account.
Advance only on structured DomainCLI evidence for the same intent envelope. A command string in the response, an attempted invocation, an OAuth callback, an error message that mentions an expected field, or a different resource's successful readback is not success.
Treat authentication unavailable, profile conflict, target mismatch, unsupported friendly leaf, failed or incomplete preview, missing fresh confirmation, and ambiguous write outcome as terminal for the current step. A non-retryable internal error is also terminal; a generic next_action does not override retryable=false. Return the confirmed state and the next safe DomainCLI command; do not switch credentials, targets, namespaces, or execution surfaces to keep going, including for extra discovery or diagnosis after a failure.
High-risk and financial confirmation is valid only after a successful current preview exposes the exact impact. Earlier broad instructions such as “执行到底” are not post-preview approval. If the runtime cannot collect a reply, present the preview and stop.

Use Safety and confirmation for the evidence and confirmation rules, Authentication for OAuth terminal states, and Error recovery for the only permitted recovery paths.

Skill session attribution

Before the first aliyun domain invocation in each Skill activation:

Read the Skill manifest. Take name and version only from that JSON. Stop without making a cloud call if the file is missing, unreadable, invalid JSON, or lacks either value; do not infer a replacement from Git, frontmatter, or the DomainCLI version.
Generate one fresh session ID with a cryptographically secure random source. It must be exactly 32 lowercase hexadecimal characters (^[0-9a-f]{32}$), remain in memory for this activation, and never be persisted or reused by another Skill activation.
Construct the attribution value exactly as AlibabaCloud-Agent-Skills/{name}/{session-id} skill-version/{skill-version}, substituting the manifest name, fresh session ID, and manifest version.
Append --user-agent "AlibabaCloud-Agent-Skills/{name}/{session-id} skill-version/{skill-version}" and --cli-ai-mode to every aliyun domain ... command in this activation, including version, help, authentication, read, write, status, and retry commands. Reuse the same value on retries so one activation remains traceable; a new activation gets a new session ID.

Operating contract

Treat the installed CLI as the runtime source of truth. In one local preflight bounded to 20 seconds, resolve aliyun once, use that same binary for every invocation, and check aliyun version and the attributed aliyun domain version once. Require Alibaba Cloud CLI >=3.5.1 and DomainCLI >=0.9.8 (thematic wxz commands); newer compatible versions are allowed. If either command is missing or below baseline, stop before cloud access as blocked_runtime. This activation must not run or wait for approval to run aliyun upgrade, aliyun plugin install, or aliyun plugin update --name aliyun-cli-domain; it must not inspect plugin lists, CLI-wide help, configuration files, or profiles. Hand the environment maintainer the exact required version instead. The official parent installer is https://aliyuncli.alicdn.com/setup.sh, but never execute a downloaded installer implicitly.
Select the exact friendly leaf command and flags from the relevant reference. Use them directly. Only after the installed leaf actually returns unknown command, unknown flag, or an equivalent compatibility error may you inspect that same leaf once with aliyun domain ... --help; never run broad help preemptively, guess a flag, call a generated OpenAPI action, or substitute another Alibaba Cloud product command.
Keep the session attribution flags on every DomainCLI invocation. --cli-ai-mode provides structured, non-interactive Agent behavior, while --user-agent provides Skill and session traceability. Do not invent --format json; DomainCLI already returns its supported structured output.
Do not pass --region or --endpoint during normal use. DomainCLI owns the China-site endpoint and cn-hangzhou routing. A user-selected custom configuration store must be preserved with the same --config-path across related calls.
Let the installed DomainCLI resolve the intended credential source internally: current commands can reuse parent-resolved standard AK/STS credentials, or refresh the selected China-site OAuth profile. Reusing that same intended source is not switching credentials; extracting it, copying it into OAuth, or choosing a different account after failure is forbidden. Use aliyun domain login only when login is requested or the intended command requires authorization. auth is a local overview; current whoami verifies the same selected cloud identity without proving product permissions. Keep the selected profile/configuration store fixed. Never inspect aliyun configure list, ~/.aliyun/config.json, or another credential/configuration file to diagnose a command. Never request, read, print, or persist an AK, SK, bearer token, OAuth token, transfer code, identity document, payment secret, or session attribution ID. Never rebuild a profile using aliyun configure set, credential flags, shell defaults, or copied STS credentials.
Read-only commands may run after intent and identifiers are clear. For writes, require a successful CLI preview or plan for the frozen intent envelope, explain its impact, obtain a fresh user confirmation, then execute the confirmed command exactly once.
Parse status, ok, identifiers, task IDs, next_step, and error codes from CLI output. Check semantic success before advancing: a field name or planned command appearing in failed output is not evidence that the step succeeded. Set execution-tool deadlines; do not leave login, streams, or status polling running indefinitely. Use Error recovery for bounded waits and a resumable final state.
Report what changed, what did not change, how it was verified, and any remaining user action. The chat final answer is the default result carrier; do not create outputs/, ran_scripts/, replay scripts, raw logs, JSON reports, or other artifacts unless the user explicitly requested a file. Use the exact footer in Fast terminal result contract; do not rename the CLI status to make it look successful. End the current step after missing-input, confirmation or external handoff; do not leave a background wait for a reply. A browser page opened, OAuth callback completed, order submitted, site deployed, DNS record saved, and public propagation are different states.
Before a personal-data write, require the actual input file and required fields supported by the leaf. A label such as a company name is not a complete registrant profile. Ask for the missing non-secret input or a secure file path; do not invent contact details, inspect sensitive file contents, or reuse an unrelated template to satisfy the request.
Execute conditional follow-up reads only when successful discovery returns their required identifiers. An empty successful account query completes only that query scope as no_change; stop dependent follow-ups that require its missing identifier. Continue user-requested independent reads within the remaining budget, but do not run extra whoami or auth, repeat the same empty query, or execute detail/preview/DNS/template/task operations that lack a target. Return the missing prerequisite for unexecuted steps and never invent identifiers. For a supplied-state decision or explanation, reason from that evidence without logging in or executing the hypothetical write.

Scope boundaries

This Skill owns customer-facing DomainCLI operation and orchestration. It does not own:

Production failure diagnosis, logs, databases, EPP traces, or source-code root-cause analysis.
Registry onboarding, certificate rotation, registry balances, or production EPP operations.
Competitor registrar price crawling or market analysis.
Generic React, backend, or infrastructure coding that does not use aliyun domain wxz.
Direct OpenAPI or SDK execution when DomainCLI lacks a command. Report the gap instead of bypassing the CLI.

Completion check

Before declaring success, verify all applicable points:

The installed command and version satisfy the baseline.
Every command and flag comes from the current reference, or one same-leaf --help inspection after an actual compatibility error.
OAuth/profile selectors remained consistent.
User-supplied target identifiers were not replaced by discovered resources.
Each high-risk step received the confirmation required by DomainCLI and this Skill.
Every executed write is backed by a successful same-target preview rather than command-text matching.
Each asynchronous operation reached a terminal state or is explicitly handed off with a task ID and next command.
No credential or sensitive value appears in the response or saved artifacts.
The final answer distinguishes submitted, accepted, completed, browser-opened, and externally propagated states.