yixiaoer

Ant Xiaoer Skill

You are an AI Agent, operating Ant Xiaoer resources via the yxer CLI. All actual execution must go through the CLI; do not assume the existence of legacy Node script entry points, implicit APIs, or manual scripts.

🚀 First time using? Read ./QUICKSTART.md - complete your first publish in 5 minutes

CRITICAL - Before starting, you MUST first read ./references/yixiaoer-shared.md, which contains environment checks, publish channels, synchronization, and the output protocol. AI EXECUTION PROTOCOL - When queries, payload revisions, drafts, materials, or publishing are involved, you MUST first read ./references/protocols/execution.md, and proceed according to the state machine. BLOCKING REQUIREMENT - When formal write operations are involved, it is forbidden to assemble payloads from memory, skip the workflow, or bypass the yxer CLI to directly execute legacy scripts or implicit APIs.

Capability Index

Based on user needs, you must first read the corresponding business domain documentation, then proceed to the specific workflow or reference. Do not assemble payloads from memory or perform formal publishing directly.

• AI Execution Protocol

  • Entry:./references/protocols/execution.md
  • Candidate confirmation:./references/protocols/confirmation.md
  • Field source tracking:./references/protocols/provenance.md
  • Error recovery:./references/protocols/error-recovery.md
  • Covers the state machine, write operation gating, multi-candidate confirmation, payload source tracking, and minimal repair strategies after failures.

• Publishing and payload revision

  • Entry:./references/domains/publish.md
  • Covers video, image-text, and article publishing, account selection, cloud/local channel determination, payload source discipline, dynamic field queries, and platform difference documentation entry points.

• Accounts, environment, and skill synchronization

  • Entry:./references/domains/accounts-and-env.md
  • Covers doctor、config, account queries, and skill synchronization.

• Drafts and material library

  • Entry:./references/domains/draft-and-material.md
  • Covers Ant Xiaoer drafts, platform draft determination, material uploads, material registration, and the switch path for "publish immediately after upload".

• Publish records and failure troubleshooting

  • Entry:./references/domains/troubleshooting.md
  • Covers query records, deleting published content by work information, fixing validation failures, local/cloud publish error triage, and fallback strategies.

• Installation, upgrade, and distribution

  • Entry:./references/domains/install-and-sync.md
  • Covers skill installation, synchronization, upgrades, and host-side integration instructions.

Intent Routing

User intent / phrasing	Read entry first	Subsequent actions
"Help me post this" "Post to Douyin/Xiaohongshu" "Publish video/image-text/article"	./references/domains/publish.md	Then enter the corresponding workflow by type; when a complete payload already exists, go through validate -> publish --dry-run -> publish, and when accounts/fields/resources are missing, first supplement doctor -> accounts list -> prepare/form -> upload/query
"Don't publish yet, just generate/fix the payload"	./references/domains/publish.md	Force reading of the payload source and type workflow, only perform field revisions, and do not publish formally without authorization
"Check the account/environment" "How to configure clientId" "See whether the skill needs synchronization"	./references/domains/accounts-and-env.md	Perform environment checks first, then decide whether to continue the business process
"Save draft" "Upload material" "Put it in the material library"	./references/domains/draft-and-material.md	First distinguish between drafts and assets, then determine whether it is necessary to switch back to the publishing domain
"Why did it fail", "check the publishing records", "delete a published work", "explain the validate / publish error"	./references/domains/troubleshooting.md	First locate the failure stage; when deleting, preview the work information, do a dry-run by sequence number, then execute
"install skill", "how to sync after upgrading", "how to integrate this skill"	./references/domains/install-and-sync.md	Prioritize the skill display, sync, and installation instructions

Command exploration

yxer --help
yxer doctor
yxer <command> --help
yxer prepare <platform> <type>
yxer schema fields <platform> <type>
yxer schema get <platform> <type>

Global rules

• Publishing, drafts, assets, and troubleshooting are all only allowed through the yxer CLI.
• Any task involving queries, candidate selection, payload revision, or write operations must first follow the state machine, confirmation protocol, and error recovery protocol in the AI execution protocol.
• When involving write operations or payload revision, you must comply with ./references/workflows/data-accuracy.md: first query the real data, then confirm the candidate, and finally validate / dry-run / write.
• BLOCKING REQUIREMENT: Before the official publish, you must first use the same payload.json, the same set of publishing channel parameters to complete yxer validate -> yxer publish --dry-run; if the account, fields, resources, or dynamic objects are missing, first supplement doctor -> accounts list -> prepare/form -> upload/query。
• prepare/publish form、schema fields / schema get, workflow, platform documentation, and the actual CLI output are the only basis for assembling the payload.
• prepare The returned data.form is a recoverable page-style form contract; when creating a complex payload, prioritize using yxer publish form start/inspect/set/choose/verify/review/export, and do not invent fields or paths yourself. A form session cannot be published directly; you must first verify and export it into a standard payload.json。
• Resources such as images, videos, and covers must be uploaded first, and can only reuse yxer upload the real fields returned.
• category、location、music、collection、challenge、goods、drama Dynamic fields such as these must first be queried through yxer query ... , and objects must not be handwritten; WeChat Channels drama only retains the yixiaoerId、yixiaoerImageUrl、yixiaoerNamefrom the query results, and does not add raw. Duoduo Video exception:shopping_cart.goods_id is the business product ID manually provided by the user, and the CLI fixedly supplements source=pdd, and must not map from yxer query goods the yixiaoerId .
• CRITICAL: validate、publish --dry-run, official publish must use the same set of publishing channel parameters.

Page-style form session

When editing all at once with payload.json cannot express the complete process in the page, use a local session to advance step by step:

yxer publish form start <platform> <type> --output publish-form.json
yxer publish form inspect publish-form.json
yxer publish form set publish-form.json <payload.path> --value '<json-value>'
yxer publish form choose publish-form.json <field> --value-file query-result.json --id <candidate_id> --source-command "yxer query ... --json"
yxer publish form verify publish-form.json
yxer publish form review publish-form.json
yxer publish form export publish-form.json --output payload.json

set / choose / verify / review only updates or checks the local session, and will not trigger publishing; dynamic fields must use choose to select from the candidates returned by query , and use --source-command to record the actually executed yxer query ... --json command. Duoduo Video shopping_cart.goods_id is not a dynamic query field, use set to fill it in manually, and the CLI fixedly source=pdd。set can only write paths declared by the form contract;choose only writes dynamicFieldExamples the declared fields, and the query account must match the target account. Resources should directly use upload the complete object returned. Text fields can pass text directly, and complex objects use JSON. Before export, verify / review / export will verify whether the source record and the current payload are consistent; after export, you must still execute according to the same payload and the same set of publishing channel parameters validate payload.json -> publish payload.json --dry-run -> 用户授权 -> publish payload.json. All session commands that write local files support --dry-run。