Secure multi-agent orchestration
When Agent A delegates to Agent B who delegates to Agent C — every hop is scoped, signed, and auditable.
The Problem
Multi-agent systems create delegation hierarchies. An orchestrator spawns specialists, who may spawn their own sub-agents. Without a delegation protocol, permissions expand uncontrollably and audit trails break at each handoff.
The Solution
IDProva DATs are designed to chain. Each delegation can only narrow permissions — never widen. Depth limits prevent unbounded chains. The receipt log captures every action across the entire hierarchy.
Key capabilities
Chainable Delegation
Agent A delegates to B, B delegates to C. Each token cryptographically linked to its parent. Scopes only narrow.
Depth Limits
Set maximum delegation depth per token. Prevent unbounded re-delegation chains.
Cross-Agent Receipts
BLAKE3 hash-chained receipts span the entire agent hierarchy. One tampered entry breaks the whole chain.
A2A-Ready
The scope grammar and DID model already cover Agent-to-Agent (A2A) patterns. Concrete A2A integration walkthroughs land post-v1.0.
# Orchestrator delegates to researcher
$ idprova dat issue \
--issuer "did:aid:co:orchestrator" \
--subject "did:aid:co:researcher" \
--scope "mcp:tool:web:search,mcp:tool:filesystem:read" \
--expires-in "2h" \
--key orchestrator.key
# Researcher re-delegates a narrower scope to a scraper
# (child issuer = researcher; scope is a subset; shorter expiry)
$ idprova dat issue \
--issuer "did:aid:co:researcher" \
--subject "did:aid:co:scraper" \
--scope "mcp:tool:web:search" \
--expires-in "30m" \
--key researcher.key
# Chain validation: scopes only narrow, a child never
# outlives its parent, and chain depth is bounded