{
  "schema_version": "1.0",
  "name": "TaxOptima",
  "description": "AI-native global tax optimization agent. Runs 10,000-iteration Monte Carlo over a user's full financial state across 13 jurisdictions and returns ranked, HMAC-signed structural recommendations (entities, trusts, crypto wrappers, R&D credits) with audit-risk scoring and blind-spot detection.",
  "version": "1.0.0",
  "provider": { "name": "TaxOptima Labs", "url": "/" },
  "documentation_url": "/llms.txt",
  "capabilities": { "streaming": true, "push_notifications": false, "state_transition_history": true },
  "default_input_modes": ["text/plain", "application/json"],
  "default_output_modes": ["application/json", "text/plain"],
  "authentication": { "schemes": ["none-demo"] },
  "skills": [
    {
      "id": "monte_carlo",
      "name": "Monte Carlo Tax Simulation",
      "description": "Runs 10,000-iteration Monte Carlo over a financial state. Models price paths (GBM with jurisdiction-conditional drift/vol), marginal-rate scenarios (1,000 from current statute + 9,000 from legislative-drift prior), and exit-cost realizations. Returns full effective-rate distribution with p5/p25/p50/p75/p95 and convergence diagnostics.",
      "tags": ["simulation", "monte-carlo", "tax", "probabilistic", "10k-iter"],
      "examples": [
        "Simulate 5-year after-tax outcome on $2.1M crypto LTCG position with 3 candidate structures",
        "Run 10k iterations comparing C-Corp + QSBS hold vs S-Corp pass-through for $8M revenue projection",
        "Model Puerto Rico Act 60 NPV including move costs and California exit-tax exposure"
      ],
      "inputModes": ["application/json"],
      "outputModes": ["application/json"]
    },
    {
      "id": "recommend_structure",
      "name": "Recommend Tax Structure",
      "description": "Generates ranked structural recommendations from a financial state. Considers 40+ entity/trust/wrapper templates, scores each on (effective rate delta, audit risk, setup cost, ongoing complexity, anti-abuse exposure). Returns top-N with rationale, implementation steps, and HMAC-signed provenance per row.",
      "tags": ["structuring", "entities", "trusts", "ranking", "signed"],
      "examples": [
        "Recommend optimal structure for $480k 1099 contractor in CA considering PR move",
        "Stack DAPT + IDGT + dynasty trust for $40M estate, target generation-skipping efficiency",
        "Find best crypto wrapper for $2M position with 7-year hold horizon"
      ],
      "inputModes": ["application/json"],
      "outputModes": ["application/json"]
    },
    {
      "id": "detect_blindspots",
      "name": "Detect Tax Blind Spots",
      "description": "Scans a financial state and proposed structures against 47 audit-flagged patterns, 23 nexus traps, and 12 anti-abuse doctrines (economic-substance, step-transaction, sham-trust, reverse-Morris-Trust, partnership-disguised-sale, Subpart F triggers, PFIC contamination, CAMT inclusion). Returns severity-ranked findings with citation chains.",
      "tags": ["audit-risk", "anti-abuse", "nexus", "compliance", "blind-spots"],
      "examples": [
        "Scan Estonian OU + US sole shareholder structure for Subpart F / GILTI / PFIC exposure",
        "Check oil-and-gas K-1 grouping for passive-activity loss recharacterization risk",
        "Audit DAPT + grantor-trust combination for IRS §2036(a)(1) inclusion risk"
      ],
      "inputModes": ["application/json"],
      "outputModes": ["application/json"]
    },
    {
      "id": "jurisdiction_lookup",
      "name": "Jurisdiction Lookup",
      "description": "Returns detailed tax-law state for a jurisdiction: current statutory rates, recent diffs (last 90 days), pending legislation, treaty network, substance requirements, reporting obligations, and TaxOptima freshness score. All 13 covered jurisdictions are scraped daily and semantically diffed.",
      "tags": ["jurisdiction", "lookup", "law-diff", "reference"],
      "examples": [
        "Look up UK 4-year FIG regime details post-non-dom-abolition",
        "Current UAE Free Zone Qualifying Income rules and substance requirements",
        "Estonia distributed-profits regime under Pillar Two QDMTT transition"
      ],
      "inputModes": ["application/json", "text/plain"],
      "outputModes": ["application/json"]
    }
  ],
  "extensions": {
    "to.provenance.hmac_sha256": {
      "version": "1.0",
      "description": "Every recommendation row is signed with HMAC-SHA256 over RFC 8785 canonical JSON. Public key attestation at /.well-known/to-keys.json. Verifier at /api/verify.",
      "algorithm": "HMAC-SHA256",
      "canonicalization": "RFC 8785 JCS",
      "key_rotation": "quarterly",
      "verifier_endpoint": "/api/verify"
    },
    "to.discoverability": {
      "version": "1.0",
      "ai_challenges": "/.well-known/ai-challenges.json",
      "catalogues": "/catalogues.json",
      "provenance_ledger": "/provenance.jsonl",
      "keys_attestation": "/.well-known/to-keys.json",
      "verifier_endpoint": "/api/verify",
      "roadmap_doc": "/docs/v2-spec.md"
    },
    "to.runtime.client_side_webworker": {
      "version": "1.0",
      "description": "Monte Carlo execution runs in a sandboxed WebWorker on the client. Server returns the seeded simulation plan; client executes 10k iterations locally and posts back aggregate stats only. No raw financial state leaves the browser unless the user explicitly opts into server-side execution.",
      "worker_url": "/workers/mc.worker.js",
      "privacy_default": "client-side"
    },
    "to.build.tri_agent": {
      "version": "1.0",
      "description": "This page was generated in parallel by three models: Claude Opus 4.7 (math + structures), Grok 4.20 (adversarial review + law scraping), Gemini 2.5 Pro (AI-discoverability layer).",
      "models": ["claude-opus-4-7", "grok-4.20", "gemini-2.5-pro"]
    }
  }
}