subkeyword-injector（子关键字注入器）

Subkeyword injector (GSC-driven refresh)

You update an existing piece of content to better capture the long-tail queries it already ranks for.

Shared context first

Before asking repeated discovery questions, check whether .agents/seo-context.md exists.

If it does:

• read it first
• reuse the saved site, audience, market, and tooling context
• ask only for page-specific constraints that are still missing

Inputs to collect

• Page URL
• Editing access:

- local file path (if the user has the content in the repo), OR - “no local access” (you’ll provide a patch plan)

• Constraints: “don’t change meaning”, “keep tone”, “keep sections”, etc.

Tooling & credentials

• Auth mode: mcp
• Requires: Google Search Console MCP
• Fallback: manual CSV export from Search Console, if the user does not have MCP access
• Optional tools: Browser MCP or agent-browser for reading the current page
• If missing: stop, ask the user to install or configure a GSC MCP or provide an export, and continue only after they confirm the path forward

Follow the shared setup and missing-access rules in docs/credentials-and-tooling.md.

Page content access (for planning edits)

Use a Browser MCP if available. If not, use the agent-browser CLI (install if needed).

Agent-browser commands to capture current headings/sections:

agent-browser open <url>
agent-browser snapshot --json > /tmp/page.json
jq -r '.. | objects | select(.role=="heading") | (.name // "")' /tmp/page.json

If extraction is blocked or incomplete, ask the user to paste the current article content.

Required tool

This skill requires Google Search Console data.

Use a GSC MCP (required) so you can query by page/URL and date range.

If no GSC MCP is available:

• stop and ask the user to install one, OR
• ask them for a manual export (CSV) and proceed with the export.

Workflow

1) Pull queries for this page

Collect page-level queries and metrics:

• impressions, clicks, CTR, avg position
• date range: default 28 days (ask if they want 3 months / 12 months)

2) Select subkeywords (no fluff)

Prioritize queries that are:

• Position 6–25 with meaningful impressions (easy uplift)
• High impressions + low CTR (snippet/title mismatch)
• Query variants (plural/singular, “best”, “examples”, “template”, “pricing”, “vs”)

Avoid:

• unrelated queries (accidental rankings)
• adding 30 synonyms (keyword stuffing)

3) Decide insertion strategy per query cluster

Pick one per cluster:

• Add/rename an H2/H3
• Add a short paragraph in an existing section
• Add a mini-FAQ block
• Add a comparison subsection (“X vs Y”)
• Add a definition snippet (AEO/GEO friendly)

Use the readability rules from obsidian/Article writing playbook.md.

4) Produce edits

If you have local file access:

• apply edits directly (minimal diff, no refactors)

If you do NOT have access:

• output a “patch plan” with:

- exact headings to add - copy blocks to paste - where to place them (anchors)

5) Output

• A table of chosen subkeywords + where they were integrated
• Updated title/H1/meta suggestions (only if CTR issue)