RuleOak Core v2.0.1

Tool Guard for governed AI actions

RuleOak Tool Guard evaluates tool calls before execution and records policy decisions, evidence, approval requests, and audit events.

Demo

A concrete governance path in one command

npm run guard:demo

search_docs: allowed
send_external_message: approval_required
delete_workspace_file: blocked

Allow

Read-only and evidence-gathering actions can proceed immediately when policy allows them.

Approval required

External, publishing, or write-like actions can be paused for human review.

Blocked

Destructive or unsafe actions can be denied before execution.

MCP Guard Pack

MCP-style tool requests can use the same RuleOak governance pattern

The v2.0 prototype converts MCP-style tool definitions into a RuleOak manifest, then evaluates proposed tool calls before execution. It does not run an MCP server or proxy live traffic yet.