Register a new guardrail that Ārai will enforce on subsequent tool calls. Use when you discover a rule mid-session that should persist for the rest of this project (e.g. 'never write to /etc', 'always run tests before push'). The rule is parsed the same way CLAUDE.md instructions are and stored locally — it takes effect on the very next PreToolUse hook.

List currently active guardrails, optionally filtered by a substring. Returns subject/predicate/object triples plus their source files so the agent can see what constraints are live before making a tool call.

Probe whether a hypothetical tool call would match any active guardrail — without executing the call or writing to the audit log. Use BEFORE taking an action you think might be regulated to avoid a deny-and-retry loop. Returns matched rules with severity (block / warn / inform) and source file:line, exactly the same shape arai_recent_decisions returns for actual firings.

Look up the most recent guardrail decisions Ārai has emitted in this session (or any session if session_id is omitted). Use this when you've just been denied or warned and want to check whether you've hit the same rule before — closes the feedback loop so you don't repeat a refused action. Returns each firing's tool, decision (deny/inject/review), the matched rule(s), and the source file the rule came from.