Update to Regit PRIIPs

{
  "name": "regit-priips-calculation",
  "description": "Calculate and review PRIIPs KID risk, scenario, and cost figures from supplied facts; compare published and fresh figures for numeric review triggers.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 288
    }
  ],
  "skill_md_contents": "---\n# SPDX-License-Identifier: LicenseRef-Regit-Proprietary\n# Copyright (c) 2026 Regit\nname: regit-priips-calculation\ndescription: Calculate and review PRIIPs KID risk, scenario, and cost figures from supplied facts; compare published and fresh figures for numeric review triggers.\n---\n\nUse this skill when the user wants to prepare or review numeric PRIIPs KID\nfigures from product facts, including a discrepancy check against an existing\nKID. Keep the job focused on the requested figures and their evidence.\n\nFirst establish that the request concerns PRIIPs, the exact product/share\nclass and figure sought, and the relevant valuation or review date, holding\nperiod, and currency context needed for that figure. For another regulation,\nstate this plugin's PRIIPs scope instead of forcing a PRIIPs operation.\n\n1. Use only the approved `https://ai.regit.io/mcp/priips` connection with its\n   exact resource and `priips:compute` scope. If linking or permission is missing,\n   explain the access step and stop before calculation.\n2. For an existing KID, use readable content available to the chat host to\n   extract the published SRI, relevant MRM/CRM classes, and moderate-scenario\n   returns by displayed holding period. Record the exact product/share class,\n   document date, units, horizons, and page or section for each figure. The MCP\n   connection does not ingest the document. Ask only when an extracted value or\n   its product context is ambiguous; if content is unreadable, request the\n   published figures and their source instead.\n3. Discover the connection's current tools and advertised input schemas. Read\n   `regit://priips/version` and `regit://priips/schemas` when available for\n   provenance and detail; the advertised tool input schema governs calls and\n   may differ structurally from the raw engine registry. Compute fresh risk,\n   cost, or scenario figures only from current supplied product data and a\n   supported tool. For SRI, obtain the MRM/CRM classes and each applicable\n   liquidity criterion rather than inferring them from a generic liquidity\n   label. For scenario presentation, use already-selected outcomes; do not\n   treat presentation arithmetic as scenario selection. Chain cost operations\n   only when their required inputs and prior Regit results exist. Never reuse\n   old KID figures as fresh recomputations. Ask for missing, ambiguous, or\n   invalid inputs, including numeric/range constraints, before calling; keep\n   decimal values as strings where the live schema requires them.\n4. If the user asks for a KID discrepancy check, compare the published and\n   freshly recomputed SRI classes and moderate-scenario returns with the\n   current numeric KID-review tool. Normalize published percentages and any\n   Regit scenario fractions to one explicitly declared return unit only with\n   deterministic exact-decimal arithmetic available to the host, never a\n   binary-float or model-estimated conversion. If that is unavailable or the\n   representation is ambiguous, request old/new values in the same stated\n   units. Align displayed holding periods; include\n   old/new MRM or CRM only as complete\n   pairs when known, and supply the benchmark/proxy applicability fact. If a\n   fresh figure cannot be computed from available current inputs, request the\n   missing data or an externally recomputed figure with its provenance.\n   Interpret shifts strictly greater than five percentage points and any\n   unmatched horizons separately. A negative numeric trigger does not settle\n   the review calendar, qualitative changes, whole-KID compliance, or revision.\n5. Present published versus fresh figures with their sources, the Regit\n   `result`, relevant `trace` steps, actual `coverage` or warnings, and engine\n   version/provenance returned. For a discrepancy, prepare candidate corrected\n   numeric figures and evidence for the manufacturer's review; do not claim to\n   generate, validate, or publish a complete KID. If a tool reports an error or\n   incomplete evidence, state that limit instead of giving an authoritative\n   number.\n"
}