autosnippet 3.2.21 → 3.3.2

package/dist/lib/service/delivery/CursorDeliveryPipeline.js

@@ -137,24 +137,27 @@ export class CursorDeliveryPipeline {
     }
     // ─── 内部方法 ───────────────────────────────────────
     /**
-     * 加载知识条目（active + high-confidence pending）
+     * 加载知识条目（active + staging + high-confidence pending）
+     *
+     * M2 六态状态机: staging/active/evolving 均为 CONSUMABLE 状态
      */
     async _loadEntries() {
         const allEntries = [];
-        // 加载 active
-        try {
-            const active = await this.knowledgeService.list({ lifecycle: 'active' }, { page: 1, pageSize: 200 });
-            const activeItems = this._extractItems(active);
-            allEntries.push(...activeItems);
-        }
-        catch (e) {
-            this.logger.warn?.(`[CursorDelivery] Failed to load active entries: ${e.message}`);
+        // 加载 active + staging + evolving（CONSUMABLE 状态）
+        for (const state of ['active', 'staging', 'evolving']) {
+            try {
+                const result = await this.knowledgeService.list({ lifecycle: state }, { page: 1, pageSize: 200 });
+                const items = this._extractItems(result);
+                allEntries.push(...items);
+            }
+            catch (e) {
+                this.logger.warn?.(`[CursorDelivery] Failed to load ${state} entries: ${e.message}`);
+            }
         }
         // 加载 pending（高置信度的也纳入）
         try {
             const pending = await this.knowledgeService.list({ lifecycle: 'pending' }, { page: 1, pageSize: 200 });
             const pendingItems = this._extractItems(pending);
-            // 过滤高置信度 pending（quality.confidence >= PENDING_MIN 或无 quality 字段）
             const highConfPending = pendingItems.filter((e) => {
                 const qual = e.quality;
                 const conf = qual?.confidence;

package/dist/lib/service/delivery/RulesGenerator.js

@@ -173,9 +173,10 @@ export class RulesGenerator {
             '',
             '- `autosnippet_bootstrap` — Cold-start: analyze the project and generate initial knowledge entries',
             '- `autosnippet_search({ query })` — Search knowledge base (BM25 + semantic)',
-            '- `autosnippet_create_recipe` — Create a new coding recipe from current context',
+            '- `autosnippet_submit_knowledge` — Submit a knowledge candidate (strict validation)',
             '- `autosnippet_guard` — Run compliance review on current changes',
-            '- `autosnippet_task_start` / `autosnippet_task_close` — Track coding tasks with automatic knowledge capture',
+            '- `autosnippet_task` — Task & decision management (prime/create/claim/close/record_decision)',
+            '- `autosnippet_panorama` — Project panorama (overview/module/gaps/health)',
             '',
             '## Recommended First Steps',
             '',

package/dist/lib/service/evolution/ConsolidationAdvisor.d.ts

@@ -0,0 +1,114 @@
+/**
+ * ConsolidationAdvisor — 提交前融合顾问
+ *
+ * 解决问题：Agent 逐条提交 Recipe 导致碎片化、低价值条目激增。
+ *
+ * 设计思路：在新知识提交前分析已有知识库，给出 4 种建议之一：
+ *   create       — 独立有价值，正常新建（走正常可信度判断）
+ *   merge        — 与 1 条 Recipe 相似，将候选内容合并到已有 Recipe，合并后 Recipe → staging
+ *   reorganize   — 与多条 Recipe 交叉重叠，将候选功能拆分到已有 Recipe 上，被修改的 Recipe → staging
+ *   insufficient — 独立价值不足且已有足够 Recipe 覆盖，交给 Agent 与开发者决定
+ *
+ * 分析维度：
+ *   1. 结构相似度 — 复用 RedundancyAnalyzer 的 4 维算法
+ *   2. 语义域覆盖 — category + trigger 是否落在已有 Recipe 管辖范围
+ *   3. 独立价值   — 内容长度、具体性、是否有独立 coreCode
+ */
+interface DatabaseLike {
+    prepare(sql: string): {
+        all(...params: unknown[]): Record<string, unknown>[];
+    };
+}
+/** 提交候选的必要字段 */
+export interface CandidateForConsolidation {
+    title: string;
+    description?: string;
+    doClause?: string;
+    dontClause?: string;
+    coreCode?: string;
+    category?: string;
+    trigger?: string;
+    whenClause?: string;
+    kind?: string;
+    content?: {
+        pattern?: string;
+        markdown?: string;
+        [key: string]: unknown;
+    };
+}
+/** 建议类型 */
+export type ConsolidationAction = 'create' | 'merge' | 'reorganize' | 'insufficient';
+/** 融合方向分析 — 描述候选能为已有 Recipe 补充什么 */
+export interface MergeDirection {
+    /** 候选提供的新维度（已有 Recipe 缺失的） */
+    addedDimensions: string[];
+    /** 融合建议摘要 */
+    summary: string;
+}
+/** 融合分析结果 */
+export interface ConsolidationAdvice {
+    action: ConsolidationAction;
+    confidence: number;
+    reason: string;
+    /** action=merge 时，将候选内容合并到的目标 Recipe */
+    targetRecipe?: {
+        id: string;
+        title: string;
+        similarity: number;
+    };
+    /** action=merge 时，候选能为目标 Recipe 补充的新维度 */
+    mergeDirection?: MergeDirection;
+    /** action=reorganize 时，需要重新组织的 Recipe 列表 */
+    reorganizeTargets?: {
+        id: string;
+        title: string;
+        similarity: number;
+    }[];
+    /** action=insufficient 时，已覆盖该领域的 Recipe */
+    coveredBy?: {
+        id: string;
+        title: string;
+        similarity: number;
+    }[];
+    /** 需要 Agent 关注的上下文 */
+    relatedRecipes?: {
+        id: string;
+        title: string;
+        similarity: number;
+    }[];
+}
+/** 批量分析结果 — 每个候选一条分析 + 批次内重叠检测 */
+export interface BatchConsolidationResult {
+    items: {
+        index: number;
+        advice: ConsolidationAdvice;
+    }[];
+    /** 批次内部候选之间的重叠 */
+    internalOverlaps: {
+        indexA: number;
+        indexB: number;
+        similarity: number;
+    }[];
+}
+export declare class ConsolidationAdvisor {
+    #private;
+    constructor(db: DatabaseLike);
+    /**
+     * 分析候选知识与现有知识库的关系，返回融合建议。
+     *
+     * @param candidate - 待提交的候选数据
+     * @returns ConsolidationAdvice — 建议 + 理由 + 上下文
+     */
+    analyze(candidate: CandidateForConsolidation): ConsolidationAdvice;
+    /**
+     * 批量分析候选知识与现有知识库的关系。
+     *
+     * 除了对每个候选独立运行 analyze() 外，
+     * 还检测批次内部候选之间的重叠（防止批量提交碎片化）。
+     *
+     * @param candidates - 待提交的候选数组
+     * @returns BatchConsolidationResult — 每条分析 + 批次内重叠
+     */
+    analyzeBatch(candidates: CandidateForConsolidation[]): BatchConsolidationResult;
+}
+export {};