@empiricalrun/test-gen 0.35.5 → 0.35.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,11 @@
1
1
  # @empiricalrun/test-gen
2
2
 
3
+ ## 0.35.6
4
+
5
+ ### Patch Changes
6
+
7
+ - 948f534: fix: update verification agent prompt
8
+
3
9
  ## 0.35.5
4
10
 
5
11
  ### Patch Changes
@@ -101,7 +101,7 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
101
101
  await actions.executeAction(toolCall.function.name, JSON.parse(toolCall.function.arguments), toolCallsSpan);
102
102
  executedActions.push({
103
103
  isError: false,
104
- action: JSON.stringify(toolCall),
104
+ action: JSON.parse(toolCall.function.arguments)?.reason,
105
105
  });
106
106
  lastActionExecTrace = "";
107
107
  }
@@ -109,7 +109,7 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
109
109
  // TODO: implement feedback loop to llm
110
110
  executedActions.push({
111
111
  isError: true,
112
- action: JSON.stringify(toolCall.function.arguments)?.reason,
112
+ action: JSON.parse(toolCall.function.arguments)?.reason,
113
113
  });
114
114
  lastActionExecTrace = e.message;
115
115
  void testgenUpdatesReporter.sendMessage(e.message);
@@ -16,7 +16,7 @@ async function verificationAgent({ trace, task, conversation, }) {
16
16
  const messages = await (0, llm_1.getPrompt)("agent-steps-verification", {
17
17
  task,
18
18
  conversation: conversation.join("\n"),
19
- }, 4);
19
+ }, 5);
20
20
  const llm = new llm_1.LLM({ provider: "openai" });
21
21
  const response = await llm.createChatCompletion({
22
22
  trace: verificationAgentSpan,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@empiricalrun/test-gen",
3
- "version": "0.35.5",
3
+ "version": "0.35.6",
4
4
  "publishConfig": {
5
5
  "registry": "https://registry.npmjs.org/",
6
6
  "access": "public"