spec(skills): sharpen model-ambiguity-handling trigger to internal-state predicate#1452
Merged
Merged
Conversation
…ate predicate
model-ambiguity-handling の auto-invoke トリガーを、表面キーワード先頭の形から「出力の確信レジスターが実際の検証基盤と一致しない」という内部状態述語へ組み替えた。verifiable なのに hedge する過小方向も新たに拾い、body の 2-step flow に忠実化。キーワード列 ("I think" / "maybe" 等) は surface tell へ降格した。
背景: #1451 のトリガー質仮説検証。本編集は brake 1 (parallel-subagent-eval N=3) で全会一致 consistent。同 PR 候補だった model-source-check の編集は、無条件 verify を calibration で条件付けて regression と判定されたため drop した (source-check の無条件トリガーはそのままが正しい、という finding)。
Refs #1451
liplus-lin-lay
commented
Jun 1, 2026
Member
Author
liplus-lin-lay
left a comment
liplus-lin-lay
left a comment
There was a problem hiding this comment.
AI self-review (mode: semi_auto / L1 → brake 2 待ち)
verdict: pass (blocking なし)。merge は Master の brake 2 レビュー待ち。
- scope:
skills/model-ambiguity-handlingの description 1 行のみ。body / 他ファイル不変。diff = 1 file, +1/-1。 - brake 1 (parallel-subagent-eval N=3): 本編集は全会一致
consistent。- coverage: 回帰なし。むしろ「verifiable なのに hedge する」過小方向を新規取得し、body の 2-step flow に忠実化。
- impression-literal: なし。
- body 整合: OLD より忠実 (register-vs-basis は body Position の言い換え)。
- EDIT 2 (model-source-check) drop: N=3 が独立に regression を指摘 —「this session」exemption が
fixed-reference-temporal-separationと衝突 /「on an unverified claim」が perfect-defense illusion を骨抜き。source-check の無条件トリガーはそのままが正しい、という finding。 - CI: governance pass。
- gate: L1-model → brake 2 (human review) 必須。
--auto不使用。 - post-merge 予定: self-eval log で application-moment misfire を before/after 観測 (post-L1-merge observation) ＋ self-evolution-observation entry。
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.
概要
skills/model-ambiguity-handlingの auto-invoke トリガー (description) を、表面キーワード先頭の形から 内部状態述語「出力の確信レジスターが実際の検証基盤と一致しない時」へ組み替える。verifiable なのに hedge する過小方向も新規に拾い、body の 2-step flow に忠実化。キーワード列は surface tell へ降格。body は不変。背景 (#1451)
トリガー質仮説の検証: 「鋭い内部状態述語化が application-moment 発火を改善するか」。本 PR は EDIT 1 (model-ambiguity-handling) のみ。
brake 1 結果 (parallel-subagent-eval N=3)
consistent→ 採用。negative×2 +partial×1 → drop。三体が独立に同じ二点を指摘:fixed-reference-temporal-separationと衝突。finding
トリガーの confidence-calibration 化は スキル依存。calibration が本分の ambiguity-handling では改善、無条件 verify が本分の source-check では regression。普遍的レバーではない。
gate
Closes #1451