360 Secure SkillHub
360 安全技能中心

MetaRun: review inbox

该 Skill 通过 MetaRun 的 MCP 工具或命令行读取应用商店评价,按低星、提问、好评排序分诊,起草公开回复并经用户逐条确认后发布或撤回,挖掘好评与投诉主题及关键词候选,并可把五星评价做成截图工作室中的社会证明幻灯片。

iderasyh
iderasyh
热度 7.65
最近更新 2026-10-07 19:56:21
分类:营销依赖:MCP 服务 / 第三方 API应用场景:文案创作 / 市场分析 / 研究 / 品牌素材

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add metarun-review-inbox
Metadata
name
review-inbox
description
Work an app's customer reviews through MetaRun: triage what needs a reply, draft and post public responses, mine reviews for product and keyword signal, and turn the best quotes into social-proof screenshot slides. Trigger phrases: "check my reviews", "reply to reviews", "what are users complaining about", "answer the 1-star reviews", "mine my reviews".
metadata
{"openclaw":{"homepage":"https://metarun.dev/agents","primaryEnv":"METARUN_TOKEN","envVars":[{"name":"METARUN_TOKEN","required":false,"description":"MetaRun personal access token (Settings, MCP). Not needed after `npx metarun login`, which stores it in ~/.metarun/config.json."}],"requires":{"anyBins":["npx"]}}}

Review inbox: triage, reply, and mine the signal

You are working customer reviews through MetaRun. Replies are PUBLIC and permanent-feeling: drafted carefully, approved explicitly, posted once.

Ground rules

  • Tools run over MetaRun's MCP server or npx metarun run <tool> '<json>'. Fetch unfamiliar schemas with npx metarun tools <tool-name> --json.
  • Posting and deleting responses requires the user's Apple key to have Admin or Customer Support role. If those tools fail on authorization, say so; do not retry blindly.
  • A reply is an outward action. Never post without showing the exact text and getting a clear yes. One review, one confirmation; no bulk-posting a template the user approved once.

The mission

1. Read the room

  • apple_list_reviews -> recent reviews with ratings and territories. Triage order: recent 1-2 star reviews with text first, then anything that asks a question, then praise worth acknowledging.
  • apple_get_review_summary -> Apple's AI summary of recent sentiment, when available.
  • Worldwide picture: aso_get_territory_ratings shows per-storefront star ratings and the weighted average, useful when the user asks "how are we doing in Japan".

2. Reply

  • Draft replies in the app's voice: specific to the complaint, no corporate boilerplate, no promises the user has not made, never defensive. If a fix shipped or is coming, say which version.
  • apple_preview_respond_to_review -> show the user the review AND the draft -> apple_apply_respond_to_review with the confirmToken.
  • A bad existing response can be withdrawn: apple_preview_delete_review_response -> apple_apply_delete_review_response.

3. Mine the signal

  • aso_mine_reviews extracts praise themes, complaint themes, and keyword candidates users actually type, without an LLM pass. Offer the user the top complaint as a roadmap signal and the keyword candidates for their metadata (staged via the localize-listing mission, never auto-applied).

4. Turn praise into assets (optional)

  • A five-star quote can become a social-proof screenshot slide: studio_add_review_slide places it in a Screenshot Studio project. Offer this when a review is short, vivid, and specific.

Beyond this mission

TestFlight crash and screenshot feedback are separate reads: discover with npx metarun tools testflight. The wider surface: npx metarun tools <filter>.