@hachej/boring-agent 0.1.33 → 0.1.35

package/dist/eval/index.js

@@ -280,47 +280,81 @@ function matchExpectations(opts, actual) {
 }
 async function runOnce(app, opts) {
   const querySuffix = formatQuerySuffix(opts.query);
-  await app.inject({
-    method: "POST",
-    url: `/api/v1/agent/sessions${querySuffix}`,
-    headers: opts.headers,
-    payload: { id: opts.sessionId }
-  });
   const userMessage = opts.systemPrompt ? `[SYSTEM]
 ${opts.systemPrompt}
 [/SYSTEM]
 
 ${opts.prompt}` : opts.prompt;
-  let res;
   try {
-    res = await withTimeout(
+    const res = await withTimeout(
       app.inject({
         method: "POST",
-        url: `/api/v1/agent/chat${querySuffix}`,
+        url: `/api/v1/agent/pi-chat/${encodeURIComponent(opts.sessionId)}/prompt${querySuffix}`,
         headers: opts.headers,
         payload: {
-          sessionId: opts.sessionId,
           message: userMessage,
+          clientNonce: `eval-${opts.sessionId}`,
           model: opts.model
         }
       }),
       opts.timeoutMs,
       `chat request exceeded ${opts.timeoutMs}ms`
     );
+    if (res.statusCode !== 202 && res.statusCode !== 200) {
+      throw new Error(
+        `Pi chat prompt returned ${res.statusCode}: ${res.body.slice(0, 256)}`
+      );
+    }
+    const snapshot = await withTimeout(
+      waitForSettledState(app, opts, querySuffix),
+      opts.timeoutMs,
+      `chat turn did not settle within ${opts.timeoutMs}ms`
+    );
+    return capturePiChatSnapshot(snapshot);
   } finally {
     void app.inject({
       method: "DELETE",
-      url: `/api/v1/agent/sessions/${encodeURIComponent(opts.sessionId)}${querySuffix}`,
+      url: `/api/v1/agent/pi-chat/sessions/${encodeURIComponent(opts.sessionId)}${querySuffix}`,
       headers: opts.headers
     }).catch(() => {
     });
   }
-  if (res.statusCode !== 200) {
-    throw new Error(
-      `chat returned ${res.statusCode}: ${res.body.slice(0, 256)}`
-    );
+}
+var STATE_POLL_INTERVAL_MS = 250;
+async function waitForSettledState(app, opts, querySuffix) {
+  for (; ; ) {
+    const state = await app.inject({
+      method: "GET",
+      url: `/api/v1/agent/pi-chat/${encodeURIComponent(opts.sessionId)}/state${querySuffix}`,
+      headers: opts.headers
+    });
+    if (state.statusCode !== 200) {
+      throw new Error(`Pi chat state returned ${state.statusCode}: ${state.body.slice(0, 256)}`);
+    }
+    const snapshot = JSON.parse(state.body);
+    if (snapshot.status !== "streaming") return snapshot;
+    await new Promise((resolve) => setTimeout(resolve, STATE_POLL_INTERVAL_MS));
+  }
+}
+function capturePiChatSnapshot(snapshot) {
+  const toolCalls = [];
+  const textParts = [];
+  const snapshotError = snapshot.error;
+  const errorText = typeof snapshotError?.message === "string" ? snapshotError.message : void 0;
+  const messages = Array.isArray(snapshot.messages) ? snapshot.messages : [];
+  for (const message of messages) {
+    if (typeof message !== "object" || message === null) continue;
+    const parts = Array.isArray(message.parts) ? message.parts : [];
+    for (const part of parts) {
+      if (typeof part !== "object" || part === null) continue;
+      const rec = part;
+      if (rec.type === "text" && typeof rec.text === "string") textParts.push(rec.text);
+      if (rec.type === "tool-call" && typeof rec.toolName === "string") {
+        toolCalls.push({ tool: rec.toolName, params: typeof rec.input === "object" && rec.input !== null ? rec.input : {} });
+      }
+    }
   }
-  return parseSseStream(res.body);
+  return { toolCalls, text: textParts.join(""), errorText };
 }
 function formatQuerySuffix(query) {
   if (!query) return "";
@@ -332,63 +366,6 @@ function formatQuerySuffix(query) {
   const encoded = params.toString();
   return encoded ? `?${encoded}` : "";
 }
-function parseSseStream(body) {
-  const toolCalls = [];
-  const textParts = [];
-  const errorParts = [];
-  let usage;
-  for (const line of body.split("\n")) {
-    const trimmed = line.trim();
-    if (!trimmed.startsWith("data:")) continue;
-    const payload = trimmed.slice("data:".length).trim();
-    if (!payload || payload === "[DONE]") continue;
-    let chunk;
-    try {
-      chunk = JSON.parse(payload);
-    } catch {
-      continue;
-    }
-    const type = chunk.type;
-    if (type === "tool-input-available") {
-      const toolName = chunk.toolName;
-      if (typeof toolName !== "string") continue;
-      const input = chunk.input;
-      const params = input && typeof input === "object" && !Array.isArray(input) ? input : {};
-      toolCalls.push({ tool: toolName, params });
-    } else if (type === "text-delta") {
-      const delta = chunk.delta;
-      if (typeof delta === "string") textParts.push(delta);
-    } else if (type === "data-usage") {
-      const data = chunk.data;
-      if (data && typeof data === "object") {
-        const obj = data;
-        if (typeof obj.input === "number" && typeof obj.output === "number") {
-          usage = { input: obj.input, output: obj.output };
-        }
-      }
-    } else if (type === "error") {
-      const text = stringifyStreamError(chunk);
-      if (text) errorParts.push(text);
-    }
-  }
-  return {
-    toolCalls,
-    text: textParts.join(""),
-    usage,
-    errorText: errorParts.length ? errorParts.join("\n") : void 0
-  };
-}
-function stringifyStreamError(chunk) {
-  for (const key of ["errorText", "message", "error"]) {
-    const value = chunk[key];
-    if (typeof value === "string" && value.trim()) return value;
-    if (value && typeof value === "object") {
-      const message = value.message;
-      if (typeof message === "string" && message.trim()) return message;
-    }
-  }
-  return "unknown stream error";
-}
 function withTimeout(p, ms, label) {
   return new Promise((resolve, reject) => {
     const timer = setTimeout(() => reject(new Error(label)), ms);