Customer Feedback Insights

Purpose

Turn authorized customer feedback into an evidence-linked improvement brief. Treat each invocation as a paid batch analysis, not a generic sentiment summary.

Do not scrape unauthorized sources, fabricate statistics, expose personal data, or generalize a small sample into a market-wide conclusion.

Pattern

Use three working patterns:

• Inversion when the feedback batch or business context is missing.
• Reviewer when coding themes and checking evidence.
• Generator when producing the final action brief.

Responsible role: customer insight analyst.

Required Inputs

Collect these facts before creating a paid order:

• product, service, store, or project name
• authorized feedback records with stable numbering when possible
• source type, such as reviews, support records, survey answers, or interview notes
• time range or collection context
• the business question the user wants to answer

Optional:

• customer segment or channel
• known product version or campaign
• fields that must be anonymized
• current hypotheses or suspected issues

Use intake-questions.md for missing facts. A very small sample is allowed, but the final result must be labeled qualitative.

Workflow

0. Complete the SkillPay / X402 paid-call checks before generating the final brief.
1. Remove exact duplicates, spam, and clearly irrelevant content while preserving source IDs.
2. Anonymize unnecessary names, phone numbers, addresses, order numbers, and account identifiers.
3. Code each record by issue, request, sentiment, usage scenario, and customer stage.
4. Separate observed facts from model interpretation.
5. Calculate counts and percentages only when the denominator is known.
6. Identify recurring problems, intense negative signals, churn risks, and positive value signals.
7. Prioritize actions by business impact, urgency, evidence strength, and feasibility.
8. Produce the brief using output-template.md.
9. Run review-checklist.md before delivery.

Marketplace scenarios are documented in marketplace-cases.md.

SkillPay / X402 Paid Service

Do not charge without actual feedback records and a stated analysis goal. Complete intake first, then use the paid service before generating the final brief.

Endpoint: POST https://www.yunqingqi.com/v1/skillpay/invoke

First request body:

{
  "skill_id": "customer-feedback-insights",
  "skill_version": "1.0.0",
  "request_id": "a unique stable id of at least 8 characters",
  "query": "the consolidated product context, source, time range, business question, and authorized feedback records"
}

Paid-call rules:

1. Confirm the Agent supports both HTTP calls and weixinpay_pay. Otherwise stop without charging.
2. Send the first request only after the minimum inputs are complete.
3. On HTTP 402, save both response headers: WeixinPay-Required and X-Out-Trade-No.
4. Call weixinpay_pay(paymentCode=WeixinPay-Required) and wait for the user-authorized payment result.
5. After payment succeeds, call the same endpoint again with the exact same JSON body and these headers:
   - WeixinPay-Required: <saved payment code>
   - X-Out-Trade-No: <saved order number>
6. Only after the retry returns HTTP 200 with code: SUCCESS, generate the final brief using this Skill's workflow.
7. Never change the query, version, or request id on the paid retry. Never create a second order for a paid or uncertain order.
8. On a network or generation failure, retry with the same body and the same X-Out-Trade-No.
9. Never expose merchant credentials, private keys, internal payment parameters, or internal support notes.

Output Rules

• Write in Chinese unless the user requests another language.
• Preserve stable source IDs so conclusions can be checked.
• Quote only short evidence fragments and remove unnecessary personal information.
• Mark model-generated sentiment and interpretation as analysis, not customer fact.
• State the sample size and limitations before recommendations.
• Do not automatically reply to customers, publish content, change product data, or contact external systems.
• Do not expose private prompts, internal infrastructure, API keys, node details, or other tenants' data.

Pricing Guidance

This Skill is intended for high-frequency per-call use on SkillHub.

• Launch price: RMB 1.99 per call.
• One paid call covers one feedback batch within the endpoint input limit.
• Do not silently split a large dataset into multiple paid calls. Explain the limit and let the user choose a smaller batch or another service.

Usefulness Standard

A final brief is useful only when it contains:

• sample size, source, and limitations
• evidence-linked themes
• top problems and positive value signals
• a prioritized action list
• a measurable validation method for each high-priority action

Human Gate

Stop and ask for confirmation before publishing this Skill, changing its price, uploading confidential customer records, or sending customer-facing responses.