88d2326c39d1144f4f62611fcd79ebe2cd418f85

Author
TheEdgeOfRage <git@theedgeofrage.com>
Committer
TheEdgeOfRage <git@theedgeofrage.com>
Date

Message

Add e2e tests for classifier checks

Diff

This diff is truncated to protect this page.

 1diff --git a/README.md b/README.md
 2index 657b66857e0ebcaf6365c1d64fff0d51e3933421..c1d723b1bea7dc6aac0927eaaebbcab57641466a 100644
 3--- a/README.md
 4+++ b/README.md
 5@@ -88,6 +88,25 @@ The threshold above passed the current live policy evaluation cases with Jev 1.1
 6 
 7diff --git a/test/core/llm.test.ts b/test/core/llm.test.ts
 8index 5d51c139b6010b22d1130fdebd3020d2354cbc99..ee93f0f6b2ed1eeed94b8ff741a69559de4d792b 100644
 9--- a/test/core/llm.test.ts
10+++ b/test/core/llm.test.ts
11@@ -50,6 +50,17 @@ async function createReviewer(): Promise<PolicyReviewer | undefined> {
12   if (!model) throw new Error("POLICY_EVAL_MODEL must be set for Pi evaluation");
13 
14   const runtime = await ModelRuntime.create();
15+  if (provider === "llama.cpp") {
16+    const providerUrl = new URL(
17+      "./extensions/llama/provider.js",
18+      import.meta.resolve("@earendil-works/pi-coding-agent"),
19+    );
20+    const { createLlamaProvider } = await import(providerUrl.href);
21+    runtime.registerNativeProvider(createLlamaProvider().provider);
22+    const refreshed = await runtime.refresh({ providers: [provider], allowNetwork: true });
23+    const error = refreshed.errors.get(provider);
24+    if (error) throw error;
25+  }
26   if (backend === "classifier") {
27     const threshold = process.env.POLICY_EVAL_ASK_THRESHOLD?.trim();
28     const askThreshold = threshold ? Number(threshold) : NaN;