SEO页面观察-行动-复盘闭环 · 创作技能

这个技能解决什么问题

以Memory为持久记录，按“读记忆→观察→决策→行动→复盘→写记忆”推进单个SEO页面；决策前强制要求3-5个领先竞品页的完整拆解，行动阶段协调建页、配图、发布前审查与IndexNow提交。适合已有产品站点与发布流程的团队，不提供数据采集与建页的具体方法。

适用任务

把一次SEO页面机会从需求观察推进到上线、提交收录与效果复盘，并把证据与结论沉淀到Memory供下一轮使用。

需要的输入

产品与站点上下文；可用的搜索/浏览器/页面检查能力；memory/run-seo-page-loop/ 目录（可为空）；目标查询族；现有发布流程与线上URL；IndexNow脚本路径 executors/indexnow/submit-indexnow.mjs 及其密钥配置。

交付内容

带日期的观察记录（含3-5个竞品页逐块拆解与信息增益缺口）；决策说明（选定动作、预期可观察结果、验证证据）；上线页面URL与发布时间；发布前审查结论与已采纳修改；IndexNow提交记录；复盘对比与下一步动作建议。

任务示例

我们做的是面向独立开发者的发票工具，站点已上线。请用 run-seo-page-loop 针对“freelance invoice template”这一查询族跑一轮：先读 memory/run-seo-page-loop/ 里已有记录，再观察当前SERP并完成3-5个领先页的逐块拆解，给出本轮唯一动作和预期结果，等我确认后再进入建页。

输出示例

（示意，非真实结果）1) Memory摘要：已记录的历史观察与未验证建议；2) 观察：目标查询族、SERP构成、3-5个领先页的搜索呈现/页面形态/可见区块/钩子/证据与负向信号/信息增益缺口；3) 决策：本轮动作=新建页面，预期可观察结果与验证证据；4) 待确认后行动：建页→配图→发布前审查→部署→IndexNow提交，记录URL与时间；5) 复盘窗口与下一步建议。

来源与使用

作者 tsingyuai；许可 Apache-2.0。已核对来源与包结构，未据此宣称完成运行实测。

查看原始来源

机器可读详情 · 读取 SKILL.md · 下载完整技能包

让 Agent 使用这个技能 ↗

技能说明全文

---
name: run-seo-page-loop
description: Run an SEO page observation-action-review loop with persistent Memory by coordinating demand research, page creation, adversarial review, image generation, IndexNow submission, and performance review. Use when taking an SEO page from opportunity discovery through publication, measurement, iteration, or continuing a previous SEO loop from Memory.
---

# Run the SEO page loop

Coordinate the loop inside the product workspace. Let the current Codex or Claude Code session control the work. Use `memory/run-seo-page-loop/` as this Model's persistent Memory.

```text
Read Memory → Observe → Decide → Act → Review → Write Memory → Next observation
```

Read [memory.md](references/memory.md) before starting. Recover relevant observations, actions, outcomes, conclusions, and next-action recommendations.

## Boundaries

- Keep this Model focused on when and why the loop moves between observation, decision, action, and review.
- Delegate data-collection methods and source-specific interpretation to Collectors.
- Delegate creation, implementation, publishing, inspection, and performance-review techniques to Executors.
- Use Runtime-native browser, search, page inspection, screenshot, and local web-testing capabilities directly.
- Add a Client only for an external API action the Runtime cannot perform natively.
- Create no fixed schema, database, dashboard, workflow state, or task queue.
- Store dated operational evidence, analysis, outcomes, and next-action recommendations in Memory.
- Apply improvements to the loop itself directly to this Model. Keep methodology-change suggestions out of Memory.

## 1. Read Memory

Read recent Memory entries and older entries relevant to the product, page, query family, or pending action. Establish what is already known, what was attempted, what happened, and which recommendation should now be tested.

## 2. Observe

Invoke `$research-seo-demand` to collect and interpret current search demand and live SERP evidence. Combine it with product context and relevant Memory.

When the loop begins from an existing page, invoke `$review-seo-performance` first to observe its current outcome.

Persist useful raw evidence and a dated observation in `memory/run-seo-page-loop/`.

## 3. Decide

Before choosing a page action, confirm that the current observation contains a competitor-page breakdown for every candidate query being considered. The breakdown must cover three to five relevant leading pages and include:

- each page's search presentation and winning page shape;
- a top-to-bottom description of its visible blocks;
- reading and conversion hooks, information density, user value, and tone;
- evidence, unique information, authorship, and negative quality signals;
- an information-gain gap synthesized across the leading pages.

Do not invoke `$create-seo-page` from keyword volume, result snippets, or a list of ranking URLs alone. When this evidence is absent, return to Observe and complete it with `$research-seo-demand`.

Choose one action supported by current evidence and historical Memory. State the expected observable result and the evidence that would confirm or challenge the decision.

Possible actions include creating a page, improving an existing page, changing its snippet, strengthening evidence, adjusting conversion, resolving discovery problems, creating a supporting page, or waiting for a defined observation window.

## 4. Act

Coordinate the relevant Executors:

1. Invoke `$create-seo-page` to design and implement the page.
2. Invoke `$generate-image` when the page needs a generated or edited asset.
3. Invoke `$review-seo-page` before release and apply accepted fixes.
4. Use the product's own checks and Runtime-native browser testing.
5. Deploy through the product's existing release process.
6. After the live URL is publicly accessible, submit it with `executors/indexnow/submit-indexnow.mjs`.

Record the action, live URL, launch time, target intent, and baseline evidence in Memory.

## 5. Review

At the appropriate observation time, invoke `$review-seo-performance`. Compare current evidence with the baseline and previous Memory. Determine whether the action improved discovery, ranking, click-through, intent fit, content usefulness, product outcomes, or AI visibility.

Invoke `$review-seo-page` again when performance evidence points to a page-quality or intent problem.

## 6. Write Memory and continue

Write the dated operational evidence, analysis, summary, outcome, and recommended next action to `memory/run-seo-page-loop/`. Link the entry to the earlier observation or action it evaluates.

When the run reveals a better loop, edit this Model's `SKILL.md` or `references/memory.md` directly. Record the real operational outcome in Memory and the improved method in the Model.

Return the selected next action to the beginning of the loop.

技能是任务材料，不会自动授予账户权限、付费授权或其他系统权限。请让 Agent 按说明核对依赖，再在授权范围内使用。