Headline feature: AgentNode Guard — runtime guardrails for AI agent
tool calls. 9 action types, configurable allow/prompt/deny policy,
strict mode, per-tool overrides, MCP argument inspection, rate
limiting, and audit integration.

Also: mitigation hint accuracy fix, version bump, CHANGELOG,
THREAT_MODEL update, README Guard section.