evalbuff: carve-based eval pipeline (delete & rebuild)#487
Merged
Conversation
New approach to evals that carves features out of the current codebase and has agents rebuild them, instead of replaying git commits. Uses OpenAI SDK (gpt-5.4) to identify and surgically remove features, then runs agents in parallel to rebuild from a natural prompt, judges against original code, and iterates on docs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Switch carve eval inner agents to Claude SDK (sonnet) with 3 parallel runs - Update carve-features to use gpt-5.4 model - Remove auto-generated discover-before-implement.md (test artifact) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
carve-features.ts: two-phase pipeline — plans carveable features across the codebase, then surgically removes each one producing diffs and ground truthrun-carve-eval.ts: runs N agents in parallel on carved repos, judges against original code, and iterates on docs using the existing doc-optimizer loopcli-init-command, agents scored 5.0 baseline → 5.5 after doc improvement, generatedpatterns/discover-before-implement.mdTest plan
npx tsc --noEmit)bun run evalbuff/src/carve-features.ts --repo . --count 3produced 3 carved featuresbun run evalbuff/src/run-carve-eval.ts --repo . --carve-file carve-2026-03-30.json --feature cli-init-command --parallelism 2ran full loop successfully🤖 Generated with Claude Code