Feature: TrustBoost MCP integration for multilingual PII sanitization in agent swarms

Ruflo v3.6 ships PII pipelines in federation — great direction.
This proposes a complementary layer for runtime input sanitization
via MCP before agent LLM calls.

The gap: regex-based PII detection misses ~30% of cases:
- Contextual PII: "call me at the usual number"
- LATAM identifiers: RFC (Mexico), CPF/CNPJ (Brazil), CUIT (Argentina)
- Japanese: マイナンバー, 運転免許証
- Compound PII: name + employer + city = identifiable

TrustBoost is an open-source MCP server that handles this gap:

POST https://api.trustboost.dev/mcp
{"tool": "sanitize_pii", "input": {"text": "agent input here"}}

→ Returns sanitized text + safety_score + risk_category

Already live on Glama:
https://glama.ai/mcp/servers/teodorofodocrispin-cmyk/trustboost-api

Integration would be a single MCP tool call before each LLM
node in the swarm — no architectural changes required.

Happy to contribute a Ruflo plugin or document the integration
pattern if direction is accepted.

No repository file, test, or entry point is named. Start by reviewing Ruflo's existing PII pipelines in federation and the proposed TrustBoost MCP endpoint, then determine where an input-sanitization call could run before each LLM node. Done means an agreed integration or plugin handles the documented multilingual cases and verifies sanitized text, safety_score, and risk_category.

Written by the indexing model from the issue text.