Add browsable INDEX.md, update README, and refine AI agent identity controls

Security/access_control_policy.md

@@ -53,7 +53,12 @@ This policy establishes mandatory requirements for granting, reviewing, and revo
 ### 3.6 Emergency / Break-Glass Access
 - Governed by [Authentication Policy §3](authentication_policy.md) (Break-glass account type) and [§2.13](authentication_policy.md) (drills).
 
-### 3.7 Enforcement
+### 3.7 AI Agent Capability Scoping
+- For AI Agent (Autonomous) accounts (per [Authentication Policy §3.1](authentication_policy.md)), RBAC role membership alone **is not sufficient** — the tool/plugin/MCP-server allow-list reachable by the agent **is** its privilege boundary and **must** be reviewed and approved with the same rigor as a role grant.  
+- A sub-agent's capability set **must** be a subset of its parent's; delegation **must not** widen scope.  
+- Agent capability grants **must** be reviewed at the same cadence as Privileged/Admin access (§3.5) given the runtime-decided, non-enumerable nature of agent action sequences.
+
+### 3.8 Enforcement
 - Violations **may** result in disciplinary action or revocation of access.  
 - Systems unable to support role-based provisioning or timely revocation **must** be remediated or formally excepted with CISO approval.