Claude-Meta-Dorker

April 21, 2026 · View on GitHub

Machine-readable reference material for constructing advanced search URLs against public websites. Intended consumer: AI agents. Human readability is a non-goal — manifests are optimised for agent parsing.

What this is

A catalogue of per-service search manifests that describe:

  1. The advanced-search URL surface of a website (query params, operators, grammar).
  2. URL construction rules an agent can follow without trial-and-error.
  3. Known constraints (rate limits, auth requirements, result caps).
  4. Provenance (when the surface was last verified, by what method).

What this is not

  • A list of Google dorks. Per-service surfaces only (GitHub search, Reddit search, HN Algolia, YouTube, arXiv, etc.).
  • A general OSINT cheatsheet for humans.
  • A crawler or scraper. Manifests are declarative; execution is the agent's problem.

Surface reconnaissance first

Before writing a manifest for a service, we run a surface recon pass (see surface-recon/) that enumerates every search-like interface the service exposes:

  • Official API (REST/GraphQL)
  • RSS / Atom feeds with query params
  • Sitemap-driven search
  • opensearch.xml descriptors
  • Public search UI URL params
  • Third-party mirrors that expose better search (e.g. HN Algolia for Hacker News)

If a non-UI surface covers the advanced-search need more reliably, the manifest points the agent there and we do not duplicate UI-dorking docs for it.

Layout

manifests/        # per-service search manifests (YAML)
schema/           # JSON Schema for manifests
agent-guides/     # agent-facing reference docs (how to consume manifests)
validation/       # Playwright-based validators that live-check URL construction
surface-recon/    # per-service reconnaissance notes feeding into manifests

Manifest shape

See schema/manifest.schema.json. Every manifest carries:

  • service, base_url, verified_at (ISO-8601 UTC)
  • surfaces[] — ranked list of search surfaces with construction grammar
  • params[] — param-level semantics (type, enum, repeatable, combinable)
  • operators[] — inline query-string operators (repo:, user:, filename:, etc.)
  • constraints — auth, rate limits, result caps, pagination
  • examples[] — canonical agent-readable URL constructions

Status

Baseline in progress. A Claude Code plugin will be released once the manifest schema and a useful set of services are stable.

Licence

MIT.