Enterprise legibility: make LucidAgentIDE discoverable to Microsoft Agent 365 Registry / Defender local-agent hooks (avoid shadow-AI classification)

[mlcyclops]

Context

Microsoft's Agent 365 (GA'd through H1 2026) now discovers third-party local agents on endpoints — their announcements explicitly name GitHub Copilot CLI, Claude Code, and Codex CLI (Agent 365 – May 2026, June 2026). Defender for Endpoint adds local-agent runtime protection (preview): prompt/tool-call inspection hooks (docs).

In an M365 E7 / Agent 365 shop, a local agent that is illegible to this stack gets classified as shadow AI and risks being blocked at the endpoint — an adoption blocker independent of product merit.

Ask (investigate, then decide scope)

1. Determine what "discoverable/legible" concretely requires: registry manifest/metadata shape, Graph API agent-registration surface, whether Entra Agent ID enrollment applies to a local desktop agent, and what Defender's agent-native hooks expect vs. its network-inspection fallback.
2. Assess compatibility with existing invariants — loopback-only (ADR-0022), per-launch capability token (ADR-0024), fail-closed gate. Defender's prompt/tool-call inspection must not become a content-egress path in CUI deployments; document the gov-mode posture explicitly (may be "network-inspection only, no agent-native hook").
3. Ship the minimal legibility increment (likely: identity/metadata manifest + documentation of Defender coexistence), with its own ADR.

Non-goals

• No dependency on Microsoft tooling for core function.
• No content exposure to the Defender hook beyond what an enterprise endpoint agent already observes.

---

Filed from the enterprise add-on's competitive review (2026-07-13); Microsoft references verified against primary sources as of that date.