Synthesis Link Research

Link Research

Purpose

Gather accurate, authoritative hyperlinks for people, organizations, and entities mentioned in blog posts, articles, presentations, or any content that references external entities.

How to Use

This skill works in two modes:

1. Interactive mode: If the user does not specify entities, ask which people and organizations need links.
2. Directed mode: If the user provides names and entities, research them directly.

People to Research

If no people are specified, ask the user which people need links. Otherwise, research the people listed.

For each person, provide context such as: role, affiliation, or why they are mentioned.

Organizations and Entities

If no organizations are specified, ask the user which organizations need links. Otherwise, research the organizations listed.

For each organization, provide context such as: what they do, where they are based, or why they are relevant.

Link Prioritization

For People

Prioritize links in this order:

1. Personal or professional website
2. Institutional or company profile page
3. LinkedIn profile
4. Other professional social media (Twitter/X, etc.)

For Organizations

Prioritize links in this order:

1. Official website
2. Primary social media presence (if no website exists)
3. Authoritative third-party profile (Wikipedia, Bloomberg, etc.) if no official presence exists

Required Output Format

Provide results in HTML format for easy copy-paste:

<!-- People -->
<a href="URL">Person Name</a>
<a href="URL">Another Person</a>

<!-- Organizations -->
<a href="URL">Organization Name</a>
<a href="URL">Another Organization</a>

If a link cannot be found for a specific entity, note that and suggest alternatives. For example: "Could not find official page for X, consider linking to their LinkedIn profile or recent interview at [URL]."

For common names or ambiguous entities, confirm the context matches before providing final links.

Tips for Best Results

1. Add context for ambiguous names. For people with common names, include distinguishing details like their company, field of expertise, or location to ensure correct identification.
2. Specify link recency. If particularly current links are needed, explicitly request the most recent official websites.
3. Request verification. For important links, provide a brief confirmation that the link is still active and matches the entity.
4. Group related entities. When writing about a specific industry or event, group requests by category to improve context.
5. Regional preferences. If region-specific versions of websites are needed, mention this requirement explicitly.

Example Use Cases

Technology Conference Recap

• Speakers: keynote speakers from various tech companies
• Organizations: conference host, sponsoring companies, featured startups

Industry Analysis

• People: key industry leaders, analysts, and innovators
• Organizations: major companies, regulatory bodies, research institutions

Personal Research

• People: authors, researchers, or experts in a field of interest
• Organizations: universities, research centers, journals, or industry associations

Dead-Link Recovery (Archive/Cleanup Mode)

When auditing an existing archive for link rot — the URL was once valid but no longer resolves — substitute in priority order. Don't take "the domain name matches a known entity" as evidence; verify.

Priority Order for Dead-URL Substitution

1. archive.org Wayback Machine archived copy of the original URL. The safest first choice. Preserves what readers would have seen at the original page. For personal-history URLs (someone's old site, a defunct consulting practice, a discontinued product page), Wayback is almost always the right answer.

2. Current canonical URL of the SAME entity (only when continuity is verifiable). Examples: mysql.org → mysql.com (same software, post-acquisition). Verify same-entity via Wayback archived content, the candidate canonical's own About page, or third-party confirmation. Do not assume "similar-looking domain" means same entity.

3. Wikipedia article for the general concept/entity (not personal/private sites). Example: a specific Yahoo Pipes URL whose pipe ID is dead → link to the Wikipedia article on Yahoo! Pipes.

4. Strip the link, keep visible text + "(no longer online)" note. Default when none of the above apply.

Wayback Machine API — Rate-Limit and Fallback

The https://archive.org/wayback/available?url=... API has aggressive rate-limiting that silently returns an empty archived-captures result for URLs that DO have captures when queried in parallel. Three-pass approach:

1. First pass — sequential queries with 1.2-second pacing
2. Second pass on the "no archive" remainder — 2-3.5-second pacing
3. Third pass — URL-form fallback for URLs the API still claims have no archive

URL-form fallback (always works if an archive exists):

curl -sS -L --max-time 15 -A "Mozilla/5.0" -o /dev/null \
  -w "%{url_effective}" "https://web.archive.org/web/2010/$ORIGINAL_URL"

If Wayback has any archive, this URL redirects to a specific timestamped archive URL matching ^/web/\d{14}/. If no archive exists, the URL stays in the unresolved /web/2010/... form. Use this to distinguish "no archive" from "API rate-limited."

LinkedIn-Specific Recovery

LinkedIn returns HTTP 999 to ALL unauthenticated automated requests, both for legacy /pub/ URLs and modern /in/ vanity URLs. Direct verification via WebFetch or curl is impossible.

Don't waste sub-agent budget on direct LinkedIn WebFetch. Use one of:

1. Google site-search (site:linkedin.com "person name" employer) — surfaces the canonical LinkedIn URL in the result set if it exists.
2. Authenticated browser session (Claude in Chrome or similar) — the user logs in once; the agent navigates and reads page content. Reliably verifies 1st-degree connection, current employer, etc.
3. Hash-suffix migration pattern — for legacy /pub/firstname-lastname/0/abc/def URLs, the modern vanity URL is often /in/firstname-lastname-defabc (the last two hex groups concatenate in reverse order). Pattern held for ~70% of cases in the 2026 rajiv-com-cleanup project; the rest used user-chosen clean vanity URLs (/in/firstname-lastname).

Browser Automation Has a Domain Allow-List

Claude in Chrome's browser session is gated per-domain. Reddit, NYT, Microsoft, GNU, SmugMug, and many other common domains return "Navigation to this domain is not allowed" until the user grants per-domain permission. For bulk URL verification across many domains, fall back to:

• curl with full browser headers (User-Agent + Accept + Accept-Language) using GET method. Bypasses HEAD-method anti-bot discrimination on most real sites.
• WebSearch for "does X site exist" or "what's the canonical URL for entity Y" — often answers without rendering the page.

URL Extraction Regex for Archive Scans

When extracting URLs from markdown/prose to feed a scan, exclude code-block / template metacharacters:

url_pattern = re.compile(r'https?://[^\s)\]'"`<\$]+')

The \$ excludes shell/template interpolations like http://$server_name. The backtick excludes URLs at the end of inline-code spans. The < excludes URLs immediately followed by HTML fragments. Without these exclusions, scans surface ~10-15% false-positive URLs from code blocks that aren't real URLs.

Related

Part of the synthesis writing craft — the writer writes, the AI assists.