skills/amazon-listing-collector/
Evidence intake商品依据收集
Builds a structured source pack and keeps unknown product facts explicit.
建立结构化来源包，对尚未确认的商品事实保持明示。
Access boundary: this implementation page and its downloadable files sit under the site's password-protected /open-source/ path. The conceptual article remains public. The linked GitHub repository is private and requires an authorized GitHub account.
访问边界：当前实现页及其下载文件位于网站需密码访问的 /open-source/ 路径下。方法文章仍然公开；链接的 GitHub 仓库为私有，需使用已授权的 GitHub 账号访问。
Implementation overview实现概览
Six stages with explicit files and review boundaries.六个阶段，每一步都有明确的文件和审核边界。
skills/amazon-review-analysis/
Review evidence评论证据分析
Extracts recurring needs, objections, contexts and customer wording from permitted review material.
从允许使用的评论材料中提取重复需求、购买顾虑、使用场景和用户用词。
skills/amazon-keyword-library-builder/
Keyword governance关键词管理
Separates discovery, intended placement and approval status instead of flattening all terms into one list.
分开记录关键词的来源、拟用位置和审核状态，而不是把所有词堆成一张表。
skills/amazon-listing-generator/
Listing draftListing 草案
Generates structured copy from approved facts and preserves claim or evidence gaps for review.
基于已确认的商品事实生成结构化文案，同时标出宣称或证据缺口。
skills/amazon-fad-initialization/
Paused ad initialization默认暂停的广告初始化
Creates a draft structure, budget plan and bulk rows that remain paused until human approval.
生成广告结构、预算计划和批量行草案，在人工审核之前始终保持暂停。
skills/amazon-ad-report-analyzer/
Report-to-action analysis从报表到待审动作
Processes one supported report per run and writes a report, action register and provenance manifest without account writes.
每次处理一份支持的报表，输出分析报告、动作表和来源清单，不回写广告账户。
Repository map仓库目录
One router, six specialized skills and one traceable example.一个总入口、六个专项 Skill，以及一套可追溯案例。
The root SKILL.md routes work to the earliest missing artifact. The food-storage-container folder is a demonstration package, not proof of live account performance.
根目录的 SKILL.md 会把任务路由到当前缺失的最早阶段。食品收纳容器目录是演示包，不是真实账户成绩证明。
amazon-listing-ads-skill-suite/
├── SKILL.md
├── README.md
├── agents/
│ └── openai.yaml
├── docs/
│ └── source-notes.md
├── scripts/
│ └── validate_suite.py
├── skills/
│ ├── amazon-listing-collector/
│ ├── amazon-review-analysis/
│ ├── amazon-keyword-library-builder/
│ ├── amazon-listing-generator/
│ ├── amazon-fad-initialization/
│ └── amazon-ad-report-analyzer/
└── examples/
└── food-storage-containers/Included draft case内含草案案例
Food-storage containers: readiness and allocation, not outcome metrics.食品收纳容器：展示准备度与预算分配，不展示虚构成绩。
Safety invariants不可绕过的安全规则
The package prepares decisions; it does not take control of the account.套件负责准备决策材料，不接管亚马逊账户。
- Evidence before inference. Unknown product facts remain unknown until a permitted source resolves them.
- 证据先于推断。未知的商品事实保持未知，直到允许使用的来源能够确认。
- Paused by default. Bulk-upload output must remain paused until marketplace IDs, budgets, bids and targeting are reviewed.
- 默认暂停。批量上传文件在站点 ID、预算、竞价和定向通过审核之前必须保持暂停。
- Human approval for exclusions. Negative-keyword suggestions stay candidates until a reviewer approves their scope and evidence.
- 否定词必须人工确认。否定词建议在审核者确认范围和证据之前一直保持为候选项。
- No performance fabrication. The sample package contains no claim of live sales, CTR, CVR, ACOS or ROAS.
- 不伪造投放表现。样例包不声称拥有真实销售、CTR、CVR、ACOS 或 ROAS 结果。
Protected delivery加密交付
Download the validated package or open the private repository.下载已验证套件，或进入私有仓库。
The website download inherits this page's password protection. GitHub access is controlled separately by repository permissions.网站下载文件继承当前页的密码保护；GitHub 则由仓库权限单独控制。