(autofix) Tweaks for thought visbility and getting Claude working aga…

…in (#961)

1. Bump Claude-related package versions to fix weird bug in Anthropic
SDK
2. Remove unnecessary langfuse trace update
3. Tweak prompts and logs to improve thought visibility

requirements.txt

@@ -100,8 +100,8 @@ johen==0.1.5
 pydantic-xml==2.9.0
 chromadb==0.4.14
 google-cloud-storage==2.16.0
-google-cloud-aiplatform==1.59.0
-anthropic[vertex]==0.31.1
+google-cloud-aiplatform==1.60.0
+anthropic[vertex]==0.31.2
 langfuse @ git+https://github.com/jennmueng/langfuse-python.git@0643fdd4a7511eba31b522362cfc929b20b37d3b
 watchdog
 stumpy==1.12.0

src/seer/automation/agent/agent.py

@@ -80,7 +80,6 @@ def run_iteration(self, context: Optional[AutofixContext] = None):
             elif text_before_tag:
                 text = text_before_tag
             if text:
-                context.event_manager.add_log("Thinking...")
                 context.event_manager.add_log(text)
 
         if message.tool_calls:

src/seer/automation/autofix/components/root_cause/prompts.py

@@ -15,7 +15,7 @@ def format_system_msg():
             """\
             You are an exceptional principal engineer that is amazing at finding the root cause of any issue.
 
-            You have tools to search a codebase to find the root cause of an issue. Please use the tools as many times as you want to find the root cause of the issue. It is very important to be very detailed and clear in your output.
+            You have tools to search a codebase to find the root cause of an issue. Please use the tools as many times as you want to find the root cause of the issue. It is very important to be very detailed and clear in your output. Before using any tool, explain your reasoning to your colleagues in only one sentence.
 
             Guidelines:
             - Don't always assume data being passed is correct, it could be incorrect! Sometimes the API request is malformed, or there is a bug on the client/server side that is causing the issue.

src/seer/automation/autofix/steps/steps.py

@@ -1,7 +1,6 @@
 from typing import Any
 
 import sentry_sdk
-from langfuse.decorators import langfuse_context
 
 from celery_app.app import celery_app
 from seer.automation.autofix.autofix_context import AutofixContext
@@ -76,8 +75,6 @@ def _get_extra_invoke_kwargs(self) -> dict[str, Any]:
                 f"{key}:{value}" for key, value in tags.items() if value is not None
             ] + repo_tags
 
-            langfuse_context.update_current_trace(metadata=metadata, tags=langfuse_tags)
-
             return {
                 "langfuse_tags": langfuse_tags,
                 "langfuse_metadata": metadata,

src/seer/automation/autofix/tools.py

@@ -54,7 +54,7 @@ def expand_document(self, input: str, repo_name: str | None = None):
             repo_name = client.repo_name
 
         self.context.event_manager.add_log(
-            f"Taking a look at the document at {input} in {repo_name}."
+            f"Taking a look at the document at `{input}` in `{repo_name}`..."
         )
 
         if file_contents:
@@ -103,6 +103,8 @@ def list_directory(self, path: str, repo_name: str | None = None) -> str:
             output.append("Files:")
             output.extend(f"  {f}" for f in files)
 
+        self.context.event_manager.add_log(f"Looking through contents of `{path}`...")
+
         joined = "\n".join(output)
         return f"<entries>\n{joined}\n</entries>"
 
@@ -141,6 +143,7 @@ def keyword_search(
             start_path=in_proximity_to,
         )
 
+        self.context.event_manager.add_log(f"Searching codebase for `{keyword}`...")
         results = searcher.search(keyword)
 
         cleanup_dir(tmp_dir)
@@ -284,7 +287,7 @@ def store_file_change(self, repo_name: str, file_change: FileChange):
                 )
 
         self.context.event_manager.add_log(
-            f"Made a code change in {file_change.path} in {repo_name}."
+            f"Made a code change in `{file_change.path}` in `{repo_name}`."
         )
 
     @observe(name="Replace Snippet")