Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

Multi-turn: criteria-derived feedback biases the agent — need a separate, criteria-blind answering prompt (incl. HITL)

オープン
  • 何か問題がありますか？
#1,228 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
38/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
静か
技術スタック
typescript
領域
ai

調査の方向性

Start by reading packages/shared/src/judge/multi-turn-loop.ts around line 561 and apps/judge/src/feedback-generator.ts, then trace how judgeResults, criteriaRegistry, and HITL tool responses flow through multi-turn runs. Done means the answering path no longer receives judge or criteria context, while also handling coding-agent requests for human input.

索引モデルが issue の本文から書いたものです。

説明

author: sinedied

Original author: @sinedied

Summary

In multi-turn runs, the feedback that steers the coding agent between iterations is derived directly from the judge criteria. This means the criteria directly influence the agent's output — the run is effectively "taught to the test," which biases/contaminates the evaluation.

Why it matters

Criteria are supposed to measure the agent, not steer it. When criteria-derived feedback is fed back as the next instruction, a criterion both defines success and tells the agent how to achieve it, inflating scores and invalidating cross-config comparisons (e.g. baseline vs. with-skills).

Current behavior (code)

  • packages/shared/src/judge/multi-turn-loop.ts:561 — nextPrompt = judgeFeedback; The coding agent's next-turn prompt is the judge feedback.
  • apps/judge/src/feedback-generator.ts — feedback is generated from judgeResults + criteriaRegistry. It masks the wording (instructed not to mention "evaluation/judge/criteria") but the content is still criteria-derived, so the steering/bias remains.
  • There is no separate prompt governing how the agent should proceed in multi-turn independent of the criteria.

Proposal

  • Introduce a separate multi-turn "answering"/continuation prompt that instructs the agent how to proceed across turns, without any access to the judge/criteria context.
  • Keep the judge/criteria strictly on the measurement side; the answering agent must be criteria-blind to avoid biasing outputs.

HITL gap

  • The same criteria-blind answering path should handle responding to the coding agent's tool response for human-in-the-loop (HITL) calls (e.g. when the agent asks a clarifying question / requests input mid-task). This doesn't appear to be handled today — the loop only feeds criteria-derived feedback as the next prompt.

Related (not duplicates)

  • #162 — feedback LLM determinism (different concern).
  • #679 — multi-turn success reporting (different concern).
主要言語
TypeScript
スター
14
フォーク
20
平均マージ
2日 12時間
マージ済み PR（30日）
28

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

microsoft/scope のほかの issue

microsoft/scope の issue をすべて見る

似ている issue

TypeScript の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。