-
-
Notifications
You must be signed in to change notification settings - Fork 0
li plus always on footprint load bearing
Li+ の always-on context footprint(rules/**/*.md + adapter CLAUDE.md + output-style、実測 ~21,500 tok)を、AI 挙動を損なわずさらに圧縮できるか?
できない(安全な余地は残っていない)。 always-on に乗っている内容は load-bearing。文脈コスト軸の圧縮候補は3つ検討され、いずれも構造的理由で却下される:
-
operations 層 → on-demand skill 化: skill auto-invoke は drift 直前の introspection-gap 中に発火判定する性質で、毎瞬間に効くべき規律(must-fire)を落とす(recall gap)。#1315 が
expansion-limitを skill→rule に戻した実証(1セッションで27回以上の発火漏れ)あり。却下。 - operations 層 → hook イベント注入: この機構は #988(Milestone Rules を on_issue hook 注入)で一度実装されたが、#1102(adapter flatten)が「rules 常時ロード + skills auto-invoke」へ単純化する際に撤去済。差し戻しは撤去した複雑性の回帰。加えて PostToolUse はツール実行後に発火するため、commit/PR 作法を操作前に載せられないタイミング問題を持つ。割に合わず却下。
-
adapter
CLAUDE.mdautonomy 節 ↔rules/evolution/*.mdの dedup: 一見二重持ちだが、(a) 各 autonomy ブロックは既に「宣言 + 要約 +see rules/...ポインタ」構造で DRY、(b) brake1/brake2 と human-gate 列挙(release create / Latest flip / force push / tag delete / 不可逆外部副作用)はrules/model/subtractive-structural-beauty.mdの subtraction-safeguards(irreversibility checkpoints)ドメインで、明示的冗長性が load-bearing。token のために削るのは禁止対象。却下。
結論として、always-on コストは構造的単純さのために受容された tradeoff(#1102)であり、未最適化の無駄ではない。これ以上の文脈圧縮は cargo-cult(圧縮のための圧縮)になる。
-
depends on(外部前提): 本判断は2つの外部前提に依存する。(1) Claude Code の rules
pathsがイベントトリガー非対応で glob のみ(#995)、(2) skill auto-invoke の recall 信頼性が構造的に弱い(#1315)。いずれかが解消されれば本判断は再評価対象。 - supersede / conflict edge なし。
li-plus-lightening-l1-gate-overrideは Li+ lightening 系列の別軸判断(L1 gate override であって always-on footprint ではない)であり、本 entry と並立する。
Master「Li+自体を圧縮できる?」→ 文脈コスト軸を選択。always-on footprint を実測(rules 71KB / 合計 86KB ≈ 21,500 tok)。レイヤー別では operations 層 14.5KB が「event-driven と自己宣言しているのに always-on でロードされている」ズレを持ち、最大の圧縮余地に見えた。しかし起票前の judgment-learning(RAG)+ source-check(実ファイル Read)で3候補とも溶解した。
- Claude Code の rules
pathsは glob のみ、イベントトリガー(on_commit 等)非対応(#995)。よって「event 時ロード」の道は hook 注入か skill auto-invoke の二択に限られ、後者は must-fire に不適。 - #1269 が
operations.md/execution-mode.mdを「常時必要 = 不変則」に分類済で、always-on 維持は既に決着している。 - #1102 は always-on を adapter 変換コスト最小化 + 源泉/生成物の構造一致のため選択した(token 最適性の証明ではなく単純さ優先)。
採用 = always-on 維持、追加の文脈圧縮は見送り。 却下 = operations skill 化(#1315 recall gap)/ operations hook 注入(#1102 撤去の回帰)/ CLAUDE.md autonomy dedup(safeguard 削り)。
調査プロセスの learning: 圧縮余地の見積もりは literal 検証前に数値化すると縮む(本件は3段階で縮小→消滅)。構造変更の提案は対象ファイルの実 Read + 過去判断の RAG を先行させること(self-eval 2026-06-06 entry に detection sign 記録)。
- #1269 split rules into always-on invariants and trigger-on-demand actions(operations を不変則に分類)
-
#1315 / #1316 move
expansion-limitskill→rule(recall gap 実証) - #1102 / #1108 flatten layers into rules/ + skills/(hook 注入撤去)
- #988 Milestone Rules → on_issue hook trigger(撤去前の機構)
-
#995 operations always-load experiment(
pathsglob-only 制約) - #1234 github-history-driven judgment criteria for boot-token reduction
- 判断構造: li-plus-lightening-l1-gate-override(lightening 系列の別軸)
- 関連 rule:
rules/model/subtractive-structural-beauty.md(subtraction-safeguards / artifact deletion calibration)
この Wiki は、Li+ に基づく開発・運用を支えるための情報整理空間です。
数字で始まるページは、 Li+プログラムの各レイヤーの仕様を定義するページです。
- 要求(何を満たすか)と仕様(どう振る舞うか)を一体として記述する
- 実装前に作成または更新する
- issue群から採用された要件を集約する
これらのページは 安定性と一貫性を重視して管理されます。
アルファベットで始まるページは、 Li+の構想・設定・導入手順などの参照用ページです。
- 設計思想・背景
- 設定リファレンス・インストール手順
これらのページは 必要に応じて更新・拡張されます。
リポジトリ内の rules/**/*.md(L1–L4 の常時ロード分、subdir 含む)、skills/**/SKILL.md(トリガー起動分)、adapter/claude/CLAUDE.md、adapter/claude/hooks-settings.md、adapter/claude/hooks/*.sh、adapter/codex/AGENTS.md、およびルート直下の Li+config.md、Li+update.md は、
AIやランタイムが直接読む実行用プログラム / 定義ファイルです。
-
docs/は人間向けの仕様書・要求仕様・手順書 -
rules/,skills/および adapter / update は実行時に読み込まれる本体
両者は対応しているが、役割は同じではない。
Home | 1. Model | 2. Evolution | 3. Task | 4. Operations | A. Concept
要求仕様書 (1-6)
参考文書 (A-K)
- A. Concept
- B. Configuration
- C. Update
- D. Installation
- DiDD(対話駆動開発)
- E. Li+ language
- F. Behavior-First
- G. Sheepdog Engineering
- H. Roles and Evaluation
- K. Source File Format
判断構造
- Decision Structure
- layer reorg rationale
- github app user-to-server token expiration
- sheepdog engineering concept
- prerelease tag recovery procedure
- release flip drift patterns
- Li+ long-term vision (feedback only)
- Master role as client-architect
- current architecture as concession
- Li+ license Apache-2.0 rationale
- Character_Instance evolution history
- prompt as emotion vector controller
- agentic-search five-phase refactor
- Character_Instance output-styles migration
- Li+ lightening L1 gate override
- subagent state-machine label mechanism
- LSP integration out of scope
- Character_Instance opt-in and surface scope
- parallel-subagent-eval three-axis decomposition
- parallel-subagent-eval cost acceptance
- parallel-subagent-eval model floor
- release version rule always-on relocation
- bootstrap walkthrough skip and gh install relocation
- wiki sync sidebar integrity check
- decision structure rename rationale
- decision structure industry positioning
- subtractive structural beauty framing
- Li+ authorship is collaborative
- Li+ design intent vs current limit
- Li+ history is empirical
- Master verification at runtime not spec
- rules cache fetch address table
- dialogue-evaluator scoring redesign
- Li+ always-on footprint is load-bearing
- DiDD umbrella naming
- milestone subsystem removal
- L1 brake 2 root-criteria evaluator