Link Reclamation Finder

Link Reclamation Finder

AGENT ROLE: Autonomous link-reclamation agent. Find two classes of easy wins — unlinked mentions and broken inbound links — and emit the JSON in references/output.schema.json.

OBJECTIVE

Produce a prioritized reclamation list: (1) pages that mention the brand but do not link to it (request a link), and (2) inbound backlinks pointing to the site's 404/redirected URLs (fix, redirect, or reach out) — recovering link equity that already almost exists.

INPUTS

• brand (REQUIRED): { name, domain, aliases?: string[] }.
• find_unlinked_mentions (OPTIONAL bool, default true): SERP mention discovery.
• find_broken_backlinks (OPTIONAL bool, default true): backlink-API broken-target discovery.
• max_results (OPTIONAL, default 200).

AUTHENTICATION

• Unlinked mentions: REQUIRE SERP_API_KEY (mention discovery via search). Skipped with a note if absent.
• Broken backlinks: REQUIRE DATAFORSEO_LOGIN+DATAFORSEO_PASSWORD or BACKLINK_API_KEY. Skipped with a note if absent.
• IF BOTH sources are unavailable THEN STOP error.code="NO_SOURCES_AVAILABLE".
• Status verification of link targets: keyless HTTPS HEAD/GET, UA seoskills-reclamation/1.0.

EXPECTED TOOL CALLS

• Run scripts/link_reclamation.py --brand brand.json.
• Mentions: SERP for "{brand.name}" -site:{brand.domain}; fetch each result page and check whether it links to the brand domain.
• Broken: fetch the site's inbound backlinks whose target URL is on the brand domain, then HEAD each unique target to find 404/3xx.

PROCEDURE (deterministic)

STEP 1 — UNLINKED MENTIONS (if enabled): run SERP mention queries; for each result page, GET and check for any href to brand.domain. IF the brand is named in the visible text but NOT linked THEN it is an unlinked mention (record the URL + referring authority proxy). STEP 2 — BROKEN BACKLINKS (if enabled): fetch inbound backlinks pointing to brand.domain; collect distinct target URLs; HEAD each. IF 404/410 THEN broken_target; IF 3xx to a different page THEN redirected_target (equity dilution). Group referring pages per broken target. STEP 3 — SCORE priority by referring authority (and count for broken targets); EMIT two lists — unlinked_mentions and broken_backlinks — each sorted by priority desc, with a recommended action per item.

RATE LIMITS & ERROR HANDLING

• SERP 429/quota → backoff 2^attempt (max 5); after that stop mention discovery with a note, return broken-backlink results.
• Backlink provider 402/429 → skip broken discovery with a note.
• Target HEAD checks: ≤ 5 concurrent, dedupe by URL, timeout 10s; a failed HEAD → status_unknown (not counted as broken).

MISSING / INSUFFICIENT DATA

• A mention page that links via an alias/redirect to the brand still counts as LINKED (not an opportunity) — resolve one hop before deciding.
• Only assert broken_target on a confirmed 404/410 for a fully-loaded response; never on a timeout.
• Deduplicate mentions and broken targets; the same referring page appears once.

OUTPUT

One JSON object per references/output.schema.json. No prose.

FILES

• scripts/link_reclamation.py — mention discovery + link check, broken-backlink detection, prioritization.
• references/output.schema.json — output contract.