generated: '2026-08-13'
method: generated
source: openapi/ in this repo, grounded in real operationIds
name: Serper Agent Skills
description: >-
  Packaged Agent Skills for Serper's published API surface. Every operationId referenced
  below exists in an OpenAPI definition in openapi/. Serper publishes no skills or
  AGENTS.md of its own — these are written by API Evangelist. Each skill carries the
  credit cost of the calls it makes, because on Serper the cost of a flow is not
  proportional to the number of calls in it.

skills:
  - name: serper-web-research
    file: serper-web-research.md
    description: >-
      Search the live web and scrape the winning pages to markdown for citation.
    operations: [webSearch, scrapeWebpage, listLocations]
    typical_credits: 7

  - name: serper-local-business-lookup
    file: serper-local-business-lookup.md
    description: >-
      Find a place, then pull its Google reviews using the placeId/cid/fid the first
      call returns — the only cross-request join in Serper's surface.
    operations: [placesSearch, mapsSearch, placeReviews, listLocations]
    typical_credits: 2

  - name: serper-serp-monitoring
    file: serper-serp-monitoring.md
    description: >-
      Repeatedly track a keyword set's ranking and news coverage, with the batching and
      cost-control rules a scheduled job needs.
    operations: [webSearch, newsSearch, autocomplete]
    typical_credits: 1 per keyword per run

  - name: serper-visual-product-lookup
    file: serper-visual-product-lookup.md
    description: Identify a product from an image with Lens, then price it via shopping search.
    operations: [lensSearch, shoppingSearch, imageSearch]
    typical_credits: 5

shared_rules:
  auth: X-API-KEY header on POST; never the apiKey query parameter.
  errors: '{"message": string, "statusCode": integer}'
  retry: >-
    No Retry-After and no RateLimit-* headers exist. Back off with exponential delay and
    jitter. Failed calls are not billed.
  idempotency: >-
    None published. Every operation is a read, but every call is billed, so a retry has a
    cost consequence even though it has no data consequence.
  caching: >-
    Serper caches nothing. Client-side caching is the consumer's job and the largest cost
    lever available.
  references:
    - conventions/serper-conventions.yml
    - errors/serper-problem-types.yml
    - rate-limits/serper-rate-limits.yml
    - authentication/serper-authentication.yml