fix: strip legacy judge messages on direct judge_config() path#174
Merged
jsonbailey merged 1 commit intomainfrom May 7, 2026
Merged
fix: strip legacy judge messages on direct judge_config() path#174jsonbailey merged 1 commit intomainfrom
jsonbailey merged 1 commit intomainfrom
Conversation
Move reserved-variable warnings and ``message_history``/
``response_to_evaluate`` marker injection from ``_create_judge_instance``
into ``_judge_config`` so both the direct ``judge_config()`` API and the
``create_judge()`` wrapper get correct legacy-message stripping.
Previously the markers were only injected by ``_create_judge_instance``,
so callers using ``judge_config()`` directly saw legacy
``{{message_history}}``/``{{response_to_evaluate}}`` placeholders
rendered to empty strings by Mustache before
``_strip_legacy_judge_messages`` ran, leaving the legacy messages intact
in the returned config.
Mirrors the JS sibling fix in launchdarkly/js-core#1364.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
keelerm84
approved these changes
May 7, 2026
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Follow-up bug fix for the judge string-input PR (#165), which has already merged.
LDAIClientexposes two ways to get a judge AI config:create_judge()->_create_judge_instance()->_judge_config()->__evaluate()judge_config()->_judge_config()->__evaluate()_judge_config()calls_strip_legacy_judge_messages()on the messages returned by__evaluate()to remove legacy template messages whose content is{{message_history}}/{{response_to_evaluate}}. The strip pass relies on those literal markers being present in the rendered content.Before this PR,
_create_judge_instance()injectedextended_variables = {..., 'message_history': '{{message_history}}', 'response_to_evaluate': '{{response_to_evaluate}}'}so the markers survived Mustache rendering. The directjudge_config()path passed user variables straight through to_judge_config(), so without the marker injection Mustache rendered the placeholders to empty strings and the strip pass found nothing matching. LegacyLDMessage(role='user', content='Evaluate: ')entries leaked into the returned config.Fix
_create_judge_instance()into_judge_config()itself, so both code paths get correct behavior._create_judge_instance()to callself._judge_config(...)directly with the user's variables.LDAIClient.judge_config()covering: legacy messages get stripped from the returned config, user variables still interpolate, and reserved variables still produce warnings.This mirrors the JS sibling fix in launchdarkly/js-core#1364.
Why a follow-up
The original PR (#165) added the strip helper and wired it into
_judge_config(), but only_create_judge_instance()injected the markers needed for the helper to detect them. That PR has already merged, so this fix is a separate follow-up rather than an amendment.Test plan
make testpasses (server-ai, langchain, openai)make lintpasses (mypy, isort, pycodestyle)client.judge_config()against a config with legacy{{message_history}}/{{response_to_evaluate}}messages and asserts they are stripped from the returned configNote
Low Risk
Low risk bug fix isolated to judge-config evaluation and warning behavior, with added regression tests to lock in legacy-message stripping and variable interpolation.
Overview
Ensures legacy judge template messages (
{{message_history}},{{response_to_evaluate}}) are stripped even when callers use the directjudge_config()API by moving reserved-variable placeholder injection (and associated warnings) into_judge_config().Simplifies judge construction to pass user variables through unchanged, and adds regression tests verifying legacy-message stripping, normal variable interpolation, and warnings when callers attempt to set reserved variable names.
Reviewed by Cursor Bugbot for commit f62ae83. Bugbot is set up for automated code reviews on this repo. Configure here.