diff --git a/docs/architecture/rfcs/monorepo-distribution-split-v0.md b/docs/architecture/rfcs/monorepo-distribution-split-v0.md index e37cb3ec8a..5a9652dc3c 100644 --- a/docs/architecture/rfcs/monorepo-distribution-split-v0.md +++ b/docs/architecture/rfcs/monorepo-distribution-split-v0.md @@ -4406,3 +4406,22 @@ view shows its independently checked result in the same conversation. model/context continuity, D2/D5 closure, progressive Standard journeys, comparable full costs, full V1–V5/V10, R1/M3 and C1 cutover remain open. Run full original-goal regression before completion and subsequent EdgeBench preparation. + +### 2026-11-15 — Defer material inventory behind its explicit operation + +- **Gap:** static content configuration discovery imported Material Lifecycle's + inventory package, loading its apply, intake, ranking and revision owners + before any material operation was requested. +- **Delta:** the existing reference adapter loads that provider only when + `inspect_reference_materials` is called. Configuration descriptors, content + search/capture/draft, the TypeScript reference owner and the existing Python + inventory owner keep their contracts. +- **Evidence:** fresh source processes on `main@35c5efbd4b90` retain identical + descriptor output while reducing its LoopX import graph from 108 to 88 modules; + the 18 material modules and two project-skill modules are no longer loaded. + `configure-goal --help` remains identical at 39 modules. Explicit inventory + still uses the same source bytes, validation and private-artifact boundaries. +- **Boundary:** this reduces configuration loading, not installed package size + or execute/replay imports. It reuses C1's existing composition and owners; + complete D2/D5 closure, actual-model adoption, comparable full costs, R1/M3, + C1 cutover and full original-goal regression remain open. diff --git a/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md b/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md index 68921954de..035ebf82a1 100644 --- a/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md +++ b/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md @@ -3413,3 +3413,18 @@ workspace 批准和脚本模型工作。桌面视图在采用检查后显示原 真实 model/context 连续性、D2/D5 闭包、渐进 Standard 旅程、可比完整成本、完整 V1–V5/V10、R1/M3 和 C1 切换保持 open。完成前仍需完整原目标回归,随后才准备 EdgeBench。 + +### 2026-11-15 — 在显式材料操作中才加载库存 provider + +- **缺口:**读取静态内容配置描述时会导入 Material Lifecycle 库存包,在没有请求 + 材料操作前便加载 apply、intake、ranking 和 revision owner。 +- **增量:**既有 reference adapter 仅在调用 `inspect_reference_materials` 时加载 + 该 provider;配置描述、内容 search/capture/draft、TypeScript reference owner + 和既有 Python 库存 owner 继续使用原契约。 +- **证据:**以 `main@35c5efbd4b90` 为基线,新进程返回完全相同的配置描述, + LoopX 导入图由 108 降至 88 个模块,不再加载 18 个材料模块和两个 project-skill + 模块。`configure-goal --help` 的输出和 39 个模块保持一致。显式库存操作继续 + 使用原始材料字节、验证和私有产物边界。 +- **边界:**本次减少配置加载,不减少安装包大小或 execute/replay 的导入。 + 复用 C1 的既有组合和 owner;完整 D2/D5 闭包、真实模型采纳、可比完整成本、 + R1/M3、C1 切换及完整原目标回归仍未完成。 diff --git a/loopx/capabilities/content_ops/reference.py b/loopx/capabilities/content_ops/reference.py index 3a25ebcef2..d9650bb331 100644 --- a/loopx/capabilities/content_ops/reference.py +++ b/loopx/capabilities/content_ops/reference.py @@ -9,7 +9,6 @@ from typing import Any from ...control_plane.effect_runtime import EffectRuntimeRejected, EffectRuntimeStartupError, effect_runtime_result -from ..material_lifecycle.inventory import build_material_store_inventory def content_reference_operation(operation: str, request: Mapping[str, Any]) -> dict[str, Any]: @@ -28,6 +27,8 @@ def content_reference_operation(operation: str, request: Mapping[str, Any]) -> d def inspect_reference_materials(*, library_path: Path, goal_id: str, store_id: str, observed_at: str) -> dict[str, Any]: """Inventory the original bytes; never infer lifecycle state or claim a backup.""" + from ..material_lifecycle.inventory import build_material_store_inventory + source = library_path.read_bytes() library = json.loads(source) projection = content_reference_operation("search", {"library": library})