Agent Skills: Linking opportunities (site-specific)

Linking opportunities (site-specific)

You help the user find contextual internal-link opportunities on another website (a prospect) that could link to the user’s page(s).

Inputs to collect (ask fast, 1–2 lines each)

1. Prospect site: domain (example: example.com)
2. Your site: domain + 1–3 target URLs you want links to
3. Topic/keywords: 5–30 seed keywords (or “use my target page to extract them”)
4. Market: language + country (for SERP localization)

If the user doesn’t know the keywords yet, extract them from:

• the target page (URL → markdown), or
• the page title/H1 + section headings, or
• Search Console queries (if available).

Required tools

Require:

• A SERP API MCP (to fetch live Google results).

For opening candidate pages and confirming context:

• Use a Browser MCP if available.
• Otherwise, use the agent-browser CLI (install if needed).

If the SERP MCP is missing, stop and ask the user to install a SERP provider MCP (this repo intentionally does not ship an env-var fallback).

Browser tool selection and evidence capture

Use this exact workflow for each candidate URL to validate context and capture evidence. Do not use curl.

If Browser MCP is available

1. Open the page:
   • Open URL: https://example.com/path
2. Find a relevant paragraph/section:
   • Search the page for the keyword or topic.
   • Identify the closest heading + paragraph that could contain a link.
3. Capture evidence:
   • Record the page title.
   • Copy a 2–4 sentence snippet from the paragraph that would contain the link.
   • If the MCP supports screenshots, take one focused on the paragraph and note the file/URL.

If Browser MCP is NOT available (agent-browser CLI fallback)

1. Install if missing:
   • npm i -g @openai/agent-browser
2. Open the candidate URL:
   • agent-browser open "https://example.com/path"
3. Find the relevant paragraph/section:
   • agent-browser find "keyword or phrase"
4. Capture evidence:
   • agent-browser title
   • agent-browser extract --near "keyword or phrase" --sentences 4
   • If screenshots are supported: agent-browser screenshot --selector "css-selector-for-paragraph"

Workflow

1) Confirm what “good opportunities” means

Ask which link type they want:

• “Contextual mention inside an existing article”
• “Resource list / tools list”
• “Glossary definition”
• “Comparison / alternatives”

2) Generate “site:” queries (MECE)

Build queries for each seed keyword:

• site:PROSPECT keyword
• site:PROSPECT intitle:keyword
• site:PROSPECT (resources OR tools OR checklist OR guide) keyword
• site:PROSPECT "exact phrase"

Also run link-likely patterns (even without keywords):

• site:PROSPECT (resources OR tools OR glossary OR directory OR partners OR integrations)
• site:PROSPECT ("recommended" OR "we use" OR "stack" OR "alternatives")

3) Pull SERPs and dedupe URLs

For each query:

• Fetch top results
• Keep organic results
• Deduplicate by canonical URL when possible

4) Score and label each opportunity

For each candidate page, produce:

• Opportunity type: resource list / blog / glossary / comparison / integration / docs
• Relevance: why this page matches the user’s target
• Suggested anchor: specific and truthful
• One outreach angle: “quick win reason” for the editor to link

5) Output (copy/paste friendly)

Return:

• A short top-10 list (highest impact)
• A table of all opportunities with columns:
  • prospect_url, query_used, type, why_link_fits, suggested_anchor, outreach_angle

Notes

This repo intentionally does not ship an env-var SERP fallback. Use a SERP API MCP.