Integrate long-term memory for build error handling#1160
Integrate long-term memory for build error handling#1160iany0 wants to merge 11 commits intogoogle:mainfrom
Conversation
|
/gcbrun exp -n wq -ag |
|
No need to worry about lint/format CI failure at the moment given the main goal is iterating on the best approach. |
|
/gcbrun exp -n wq -ag |
|
/gcbrun exp -n wq -ag -rc |
|
/gcbrun exp -n wq -ag -rag |
|
/gcbrun exp -n wq-rag -ag -- -rag |
|
/gcbrun exp -n wq -m vertex_ai_gemini-2-5-pro-chat -ag |
|
/gcbrun exp -n wqrag -m vertex_ai_gemini-2-5-pro-chat -ag -- -rag |
|
/gcbrun exp -n wqrag -m vertex_ai_gemini-2-5-pro-chat -ag |
|
/gcbrun exp -n wqragg -m vertex_ai_gemini-2-5-pro-chat -ag |
|
Requested GKE Job: ofg-pr-1160-wqragg JOB: https://console.cloud.google.com/kubernetes/job/us-central1-c/llm-experiment/default/ofg-pr-1160-wqragg |
3 participants
This PR brings RAG-based long-term memory into the Prototyper for build error handling.
First, the Prototyper now has new prompt templates and an updated workflow so it can pull in guidance from stored error knowledge when dealing with build failures.
Second, under the
helperfolder there are two additions:error_classifier.py– to categorize build errors.error_patterns.yaml– a small database of regex patterns with “good” and “bad” strategies, built from past fixes.With this setup the Prototyper can match new errors against known patterns, retrieve useful hints, and reuse past reasoning to handle recurring build issues.