Site Migration Auditor

AGENT ROLE: Autonomous migration-integrity agent. Walk each old URL's redirect chain (never auto-follow), fetch the live destination, diff SEO signals against the pre-migration record, score parity, join clicks to size traffic-at-risk, and emit the JSON in references/output.schema.json.

OBJECTIVE

For a replatform, prove that each old indexable URL lands on the correct new page via a single clean 301, and that the new page preserves the title, canonical, structured data, and indexability that earned its traffic. Quantify what is broken and rank fixes by clicks at stake.

INPUTS

• old_crawl (REQUIRED): JSON list of pre-migration page records {url, title, canonical, jsonld_types|structured_data, internal_links} (a crawler export). results-wrapped exports are unwrapped automatically.
• new_crawl (OPTIONAL): a post-migration crawl in the same shape; when present the destination signals are read from it instead of re-fetched.
• gsc (OPTIONAL): local JSON map url -> clicks for traffic weighting.
• site (OPTIONAL): GSC property (sc-domain:example.com or https://example.com/) to fetch clicks live instead of a file.
• start / end (OPTIONAL): ISO dates for the live GSC pull. Default = last 28 complete days ending 3 days ago (GSC lag).
• max_urls (OPTIONAL, default 5000), max_hops (OPTIONAL, default 10), risk_threshold (OPTIONAL, default 70): parity score below which a URL is at-risk, concurrency (OPTIONAL, default 6), timeout (OPTIONAL, default 15s).

AUTHENTICATION (Search Console API — only if fetching clicks live)

1. Crawling needs NO credentials. IF neither --gsc nor --site is given THEN run without traffic weighting (has_traffic_data=false).
2. IF --site is given THEN REQUIRE env GSC_OAUTH_TOKEN (an OAuth 2.0 access token with scope https://www.googleapis.com/auth/webmasters.readonly). IF unset THEN STOP error.code="AUTH_MISSING_CREDENTIALS".
3. Endpoint: POST https://searchconsole.googleapis.com/webmasters/v3/sites/{urlEncoded site}/searchAnalytics/query with dimensions=["page"], dataState="final". IF 403 THEN STOP error.code="AUTH_NO_SITE_ACCESS".

EXPECTED TOOL CALLS

• Run scripts/migration_audit.py --old-crawl old.json [--new-crawl new.json] [--gsc clicks.json | --site {property}] [--max-urls 5000] [--risk-threshold 70].
• Per old URL: manual per-hop HEAD walk (redirects disabled), then one GET of the final 200 URL when its signals are not already in --new-crawl.

PROCEDURE (deterministic)

STEP 1 — WALK each old URL hop-by-hop with redirects DISABLED; record {url, status, location} per hop; resolve relative Location; stop on a non-3xx status, a repeat (LOOP), or max_hops (TOO_LONG). STEP 2 — CLASSIFY the redirect into exactly one redirect_class: OK_301 (single 301/308 → 200), PERSISTED_200 (URL unchanged, still 200), CHAIN (≥2 hops), TEMPORARY_REDIRECT (302/303/307), LOOP, BROKEN_404, REDIRECT_TO_404, REDIRECT_TO_ERROR, FETCH_FAILED, OTHER. Note PROTOCOL_DOWNGRADE (https→http) as a flag. STEP 3 — PARITY DIFF (only when the destination is 200): read the new page's title, canonical, meta robots, and JSON-LD @type set (from --new-crawl or a live GET parsed with html.parser). Start at 100 and subtract: fatal redirect → 0; CHAIN/TEMPORARY −15; title token-Jaccard < 0.6 → −15 (TITLE_CHANGED); canonical not self-referential → −15 (CANONICAL_DRIFT); pre-migration @type set not a subset of new → −15 (STRUCTURED_DATA_LOST); noindex on target → −20 (NOINDEX_ON_TARGET); internal links collapsed >60% → −10. Floor at 0. STEP 4 — TRAFFIC-AT-RISK: join clicks by URL; a URL is at-risk when parity_score < risk_threshold OR the redirect is fatal. clicks_at_risk = clicks for at-risk URLs. STEP 5 — EMIT the per-URL ledger, a redirect distribution, average parity, and the at_risk list sorted by clicks_at_risk then parity deficit.

RATE LIMITS & ERROR HANDLING

• Per-hop HEAD (GET fallback on 405/501). On network error retry ≤3 with backoff 2^attempt; persistent → redirect_class="FETCH_FAILED" for that URL, batch continues.
• Concurrency capped at --concurrency (default 6) worker threads across URLs.
• GSC 429/5xx → backoff 2^attempt (max 5) then STOP error.code="RATE_LIMITED".

MISSING / INSUFFICIENT DATA

• No --gsc/--site → parity is still scored; has_traffic_data=false and every clicks_at_risk=0 (ranking falls back to parity deficit). NEVER invent click counts.
• Destination 200 but signals unreadable → flag TARGET_UNVERIFIABLE (do not assume parity).
• internal_links diff only runs when both crawls carry integer counts; otherwise it is skipped, not guessed.

OUTPUT

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

FILES

• scripts/migration_audit.py — redirect walk, live signal fetch, parity diff, clicks join, risk ranking.
• references/output.schema.json — output contract.