Parrot diet and health questions answered from memory instead of calling tools - #4
Draft
langsmith-engine[bot] wants to merge 1 commit into
Draft
Conversation
The parrot-expert system prompt previously allowed the model to answer parrot diet/food-safety and health/care questions from memory, which silently bypassed get_diet_advice and get_care_tips on a fraction of identical inputs (e.g. "Is chocolate safe for my parrot?"). Add explicit mandatory-tool-use rules for diet/food-safety, health/housing/ enrichment/socialization, and species-specific questions so the safety-verification path is never skipped.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The agent sometimes answers parrot diet-safety and health questions without calling any tool, bypassing the grounded-data path: "Is chocolate safe for my parrot?" was answered from memory in trace 1 and trace 2 (other identical questions correctly call
get_diet_advice), and "What are the signs that my parrot might be sick?" was answered from memory in trace 3 (sibling traces correctly callget_care_tips(topic="health")). The behavior is inconsistent across identical inputs, and thefood_safety_onlineevaluator scores 0 on the chocolate-from-memory case. Even if the model's memorized answer is correct, the safety-verification path is being silently skipped.