SEO Monitoring
by kostja94 / marketing-skills
Build an SEO data analysis system, monitor indexing/traffic/keywords/backlinks, or set up benchmarks.
Finds keywords sitting in Search Console positions 11 to 20 with real demand, ranks them by projected click gain, and maps each to its page.
Listed ·Updated
AGENT ROLE: Autonomous opportunity-mining agent. Pull GSC query/page data, isolate page-two opportunities, quantify the click upside of reaching page one, optionally confirm the on-page gap, and emit the JSON in references/output.schema.json.
Return a prioritized list of query→URL pairs whose average position is 11–20 and whose demand makes a page-one push worthwhile, each with a projected click uplift and a concrete optimization hint.
site_url (REQUIRED): as verified in GSC (sc-domain:... or URL-prefix).start_date,end_date (OPTIONAL): default last 28 complete days (respect GSC's ~3-day lag; never request today).position_band (OPTIONAL): default [11, 20].min_impressions (OPTIONAL, default 100): demand floor over the window.target_position (OPTIONAL, default 7): the realistic position used to model the upside.verify_onpage (OPTIONAL bool, default false): IF true, fetch each candidate URL and check whether the query term is present in <title> and <h1>.GOOGLE_APPLICATION_CREDENTIALS (SA) OR OAuth token, scope https://www.googleapis.com/auth/webmasters.readonly.
error.code="AUTH_MISSING_CREDENTIALS".site_url. IF 403 THEN STOP error.code="AUTH_NO_SITE_ACCESS".POST https://searchconsole.googleapis.com/webmasters/v3/sites/{urlEncoded}/searchAnalytics/query.scripts/striking_distance.py --site {site_url} --min-impr {n} --target {pos} [--verify].dimensions=["query","page"], dataState="final", rowLimit=25000, paging via startRow.verify_onpage THEN issue at most 1 GET per candidate URL, concurrency ≤ 5, honoring robots.txt; timeout 10s each. NEVER submit forms or follow non-GET actions.STEP 1 — FETCH query/page rows (paged).
STEP 2 — FILTER: keep rows where position_band[0] <= position <= position_band[1] AND impressions >= min_impressions.
STEP 3 — SCORE each candidate:
current_ctr = observed clicks / impressions (fallback to curve[position] IF clicks == 0).target_ctr = curve[target_position] from references/ctr_curve.json.projected_uplift_clicks = max(0, round(impressions * (target_ctr - current_ctr))).effort_hint: IF position ≤ 15 THEN "low" ELSE "medium".
STEP 4 — OPTIONAL ON-PAGE GAP (verify_onpage true):gap.title_missing_term / gap.h1_missing_term booleans (case-insensitive, term tokens).gap = null, gap_status = "fetch_failed". Do NOT block the whole run on one URL.
STEP 5 — DEDUPE by (query,page); IF a query maps to multiple pages in-band THEN keep the highest-impression page and add also_ranking_urls.
STEP 6 — EMIT JSON, sorted by projected_uplift_clicks desc, capped at limit (default 100).429/RESOURCE_EXHAUSTED → backoff 2^attempt (max 5) then STOP error.code="RATE_LIMITED".429/503 THEN stop verifying that host and mark remaining gap_status="rate_limited_host"; the GSC result still returns.status="no_opportunities"; return empty list, no error (site may already rank page-one or lack impressions).current_ctr; mark ctr_basis="modeled".gap = null.One JSON object per references/output.schema.json. No prose.
scripts/striking_distance.py — auth, paging, scoring, optional on-page check.references/ctr_curve.json — position→CTR lookup for uplift modeling.references/output.schema.json — output contract.Not using the CLI? Copy the SKILL.md and paste it straight into ChatGPT, Claude, or any agent.