Conversation
c577fc4 to
cb57aa5
Compare
5e67cbe to
a12e781
Compare
55d9bdf to
d898c24
Compare
030e036 to
9dbd5ee
Compare
d898c24 to
53ba089
Compare
9dbd5ee to
05c0d25
Compare
05c0d25 to
dff3cae
Compare
5bb8023 to
cc8ea0b
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cc8ea0b6ee
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
cc8ea0b to
b2aaf54
Compare
b2aaf54 to
eccb558
Compare
eccb558 to
758f52b
Compare
|
@BugBot review |
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
| .get(evaluation_name) | ||
| .ok_or_else(|| { | ||
| Error::new(ErrorDetails::Config { | ||
| message: format!("Evaluation '{evaluation_name}' not found in config"), |
There was a problem hiding this comment.
Error message uses single quotes instead of backticks
Low Severity
The newly added error message wraps evaluation_name in single quotes (') instead of backticks. This is inconsistent with the same error message at line 659 in validate_gepa_config_uninitialized, which correctly uses backticks. This violates the rule requiring backticks to wrap technical terms in error messages.
Triggered by team rule: Prefer backticks (`) instead of ticks (') to wrap technical terms


Step towards #7011.
This creates new Stored optimizer config types so we can continue deserializing old Gepa configs with evaluation names, and creates a new gepa config option that takes function_name and evaluator_names.
Note
Medium Risk
Touches GEPA optimizer configuration, validation, and evaluation execution paths; mistakes could break optimizer launches or change which evaluators are run. Backward-compat snapshot handling reduces rollout risk but adds new config-state permutations to test.
Overview
GEPA is updated to accept either a deprecated top-level
evaluation_nameor new function-scopedevaluator_names, enabling GEPA to run directly against evaluators defined under[functions.<fn>.evaluators].This threads the new fields through durable GEPA step params/checkpoints and runtime evaluation calls, builds inline evaluation configs from function evaluators when
evaluator_namesis used, and adjusts dataset labeling accordingly. Core config loading adds stored optimizer snapshot types to preserve historical GEPA configs and introduces validation + deprecation warnings (with new tests) to enforce exactly one ofevaluation_name/evaluator_names.Written by Cursor Bugbot for commit 758f52b. This will update automatically on new commits. Configure here.