Per-trust-level policies (BLOCK, REDACT, HASH, PASS) on data flow. What about the actions themselves?

[javmann100]

I build mnki, an open trust layer for agents: identity, delegated authority with constraints, and a signed evidence list per decision, sitting underneath MCP and A2A. Swarms are the case where delegation between agents stops being theory, which is why I would like your view.

Your PII-gated data flow with per-trust-level policies, the audit trail on both sides, and the cost tracker with budgets are the closest thing I have seen in a harness to what I am building for actions: this agent, acting for this principal, may call this tool with these limits, and here is the record. When a swarm hands work to a sub-agent today, what carries the authority: the whole parent's, a narrowed slice, or nothing explicit? And do users ask for it to be narrowed?

Twenty minutes would help me; a reply here is fine too. Reference implementation, Apache-2.0: https://github.com/MNKIAgentOS/agent-trust; integration paths: https://mnki.com/docs/integrations. Thank you for ruflo; the background workers that audit and test on their own are a genuinely new idea.