feat: de-risk agent loop — semantic tool schema + agent loop + real-WP spike harness

- api/agents: tool-schemas (semantic allowlist), agent-loop (multi-step + budget), circuit-breaker, llm-client/tool-executor interfaces
- api/agents: WpRestExecutor (real WP REST), OpenRouterLlmClient
- e2e/agent: multi-step prompts + run-agent-spike runner (scores rendered changes)
- ADR 0017; 101 unit tests green

api/__tests__/agents/circuit-breaker.test.ts

@@ -0,0 +1,25 @@
+import { describe, it, expect } from 'vitest';
+import { CircuitBreaker } from '../../src/agents/circuit-breaker.ts';
+
+describe('CircuitBreaker', () => {
+  it('halts after two consecutive failures', () => {
+    const breaker = new CircuitBreaker({ maxConsecutiveFailures: 2 });
+    breaker.recordFailure();
+    breaker.recordFailure();
+    expect(breaker.shouldHalt()).toBe(true);
+  });
+
+  it('does not halt after a single failure', () => {
+    const breaker = new CircuitBreaker({ maxConsecutiveFailures: 2 });
+    breaker.recordFailure();
+    expect(breaker.shouldHalt()).toBe(false);
+  });
+
+  it('resets the failure count after a success', () => {
+    const breaker = new CircuitBreaker({ maxConsecutiveFailures: 2 });
+    breaker.recordFailure();
+    breaker.recordSuccess();
+    breaker.recordFailure();
+    expect(breaker.shouldHalt()).toBe(false);
+  });
+});