AGT v3.5 Brings Enterprise Identity, Cost Kill Switches, and Multi-Agent Collective Policies

Tools

AGT v3.5.0: Citadel Integration, Cost Governance, and Multi-Agent Collective Policies

The biggest AGT release yet ships Entra identity bridging via APIM, tiered cost budgets with auto-throttle and kill switches, and policy evaluation that spans entire agent workflows instead of individual agents. If you're running multi-agent orchestration in prod, the collective policy evaluator and intent-based authorization lifecycle are the two things worth reading first.

Decision BOM Reconstructible View: Full Audit Lineage for Any Agent Action

AGT v3.5 adds a Decision BOM (Bill of Materials) view that reconstructs the full decision lineage for any agent action from observability signals, with partial reconstruction fallback when sources are unavailable. This is the missing piece for post-incident forensics in agentic systems.

Agent Threat Rules v2.0 Ships Public Blacklist, Whitelist, and Full Skill Scan Data

ATR v2.0 is a major structural release adding a public threat feed with 1302 blacklisted skills, an ATR Verified whitelist, and a full 101K skill scan dataset from Skills.sh and OpenClaw. The RFC-001 quality standard upgrade of 94 out of 109 rules and a new GitHub Action for CI integration make this the most production-ready version of the open detection standard yet.

Analysis

AGT v3.1.0 Showcase Retrospective: What Shipped and What's Next

The closed showcase issue documents the full v3.1.0 post-launch wave including Data-Layer ABAC, AGT Lite zero-config governance in 3 lines, and a reversibility checker that assesses agent actions before execution. Good reference for teams evaluating how far the governance stack has come since the initial Microsoft-signed preview.