dsh-continual-evolve 0.15.6 → 0.15.8

package/lib/memory-rules.d.ts

@@ -38,6 +38,13 @@ export interface MemoryProposal {
     body?: string | undefined;
     /** Trajectory provenance for extraction writes ("820-831"). */
     sourceSeqs?: string | undefined;
+    /**
+     * Optional id of the same-workspace memory this proposal corrects or
+     * supersedes. The gate requires the target to exist and be active and
+     * forbids self-links; the link persists on the row for future
+     * correction-chasing. Append/replace only — no unlink path.
+     */
+    corrects?: string | undefined;
 }
 /** The merged field values a create/update would write (delete checks none). */
 export interface ProposedFields {

package/lib/memory-snapshot.d.ts

@@ -1,4 +1,4 @@
-import { MEMORY_FILE_PREVIEW_LIMIT, type MemoryStore, type ReviewSummary } from "./store.js";
+import { MEMORY_FILE_PREVIEW_LIMIT, type MemoryStore } from "./store.js";
 export { MEMORY_FILE_PREVIEW_LIMIT };
 /** One memory row the card renders. */
 export interface MemoryFileInfo {
@@ -32,8 +32,6 @@ export interface MemorySnapshot {
         orphanFtsRows: number;
         quarantined: number;
     } | null;
-    /** Review-loop audit over the workspace's extraction ledger (zeros when none). */
-    review: ReviewSummary;
 }
 /**
  * The read-only snapshot for one workspace partition. A workspace without

package/lib/memory-snapshot.js

@@ -8,7 +8,7 @@
  */
 import { existsSync } from "node:fs";
 import { basename } from "node:path";
-import { MEMORY_FILE_PREVIEW_LIMIT, zeroReviewSummary } from "./store.js";
+import { MEMORY_FILE_PREVIEW_LIMIT } from "./store.js";
 import { skillUsagePartition } from "./skill-rules.js";
 export { MEMORY_FILE_PREVIEW_LIMIT };
 /**
@@ -27,7 +27,6 @@ export function memorySnapshot(store, root) {
         files: [],
         quarantined: [],
         lastPatrol: null,
-        review: zeroReviewSummary(),
     };
     let all;
     try {
@@ -66,17 +65,8 @@ export function memorySnapshot(store, root) {
             .filter((record) => record.status === "quarantined")
             .map((record) => ({ id: record.id, reason: quarantineReasonOf(store, record) })),
         lastPatrol,
-        review: reviewOf(store, workspaceRoot),
     };
 }
-function reviewOf(store, root) {
-    try {
-        return store.reviewSummary(root);
-    }
-    catch {
-        return zeroReviewSummary();
-    }
-}
 function infoOf(record) {
     return {
         id: record.id,

package/lib/memory-tools.js

@@ -106,7 +106,14 @@ export function registerMemoryTools(tools, store) {
         const limit = clampLimit(args.limit);
         if (typeof args.id === "string" && args.id.length > 0) {
             const record = store.get(cwd, args.id);
-            return record === undefined ? `没有找到 id 为 "${args.id}" 的记忆。` : formatRecord(record);
+            if (record === undefined) {
+                return `没有找到 id 为 "${args.id}" 的记忆。`;
+            }
+            // Presence heartbeat (mirrors skill_read): a body served
+            // by id means the model opened it for the task at hand.
+            // Search/list hits are noise and stay uncounted.
+            store.recordMemoryUse(cwd, record.id);
+            return formatRecord(record);
         }
         const query = typeof args.query === "string" ? args.query : "";
         const records = query.length > 0 ? store.search(cwd, query, limit) : store.list(cwd).slice(0, limit);

package/lib/review-prompt.d.ts

@@ -0,0 +1,55 @@
+/**
+ * The content-quality review prompts (the model side of the autonomous
+ * evolution loop): a cheap gate that decides whether an item deserves a
+ * close look, and the evolve pass that judges content quality dimension by
+ * dimension and proposes the rewrite. Code owns every consequence — the
+ * model only returns verdict JSON, never touches the store.
+ *
+ * Memories are judged on five dimensions (accurate, usable, reusable,
+ * non-duplicate, fresh); skills on four (description trigger precision,
+ * body executability, reusability, non-duplicate). The prompts are
+ * deliberately behavior-blind: there are no task datasets here (unlike
+ * Hermes GEPA), only the item text plus its provenance.
+ */
+export declare const REVIEW_CHEAP_SYSTEM_PROMPT = "\u4F60\u662F\u8BB0\u5FC6\u5E93\u7684\u5185\u5BB9\u8BC4\u5BA1\u95E8\u3002\u5148\u8FC7\u5F52\u5C5E\u4E09\u6D4B\uFF1A\u2460\u4E3B\u8BED\u662F\u7528\u6237/\u73AF\u5883\u8FD8\u662F\u67D0\u4ED3\u5E93\u7684\u67D0\u7248\u673A\u5236\uFF08\u7248\u672C\u53F7\u3001\u9608\u503C\u3001\u5F00\u5173\u540D\u51FA\u73B0\u5373\u540E\u8005\uFF09\uFF1B\u2461\u62B9\u6389\u4ED3\u5E93\u540D\u7248\u672C\u53F7\u673A\u5236\u540D\u540E\u53E5\u5B50\u8FD8\u6210\u7ACB\u5417\uFF1B\u2462\u4ED3\u5E93\u4E8B\u5B9E\u4F4F ADR/docs\uFF0C\u8BB0\u5FC6\u53EA\u6536\u8DDF\u4EBA\u8D70\u7684\u3002\u5F52\u5C5E\u4E0D\u8FC7\u76F4\u63A5\u5224 false \u5E76\u6CE8\u660E\u653E\u9519\u5BB6\u3002\u518D\u5224\u503C\u4E0D\u503C\u5F97\u7EC6\u770B\uFF1A\u6709\u660E\u786E\u4E8B\u5B9E\u3001\u505A\u6CD5\u3001\u504F\u597D\u6216\u53EF\u590D\u7528\u89C4\u7A0B\u5C31\u6279\uFF1B\u7A7A\u8BDD\u3001\u7EAF\u8FC7\u7A0B\u8BB0\u5F55\u3001\u4E00\u6B21\u6027\u7684\u77AC\u65F6\u8F93\u51FA\u3001\u5F53\u8F6E\u5DF2\u53EF\u63A8\u5BFC\u7684\u4E00\u5F8B\u62D2\u3002\u53EA\u8F93\u51FA JSON\uFF1A{\"worthReview\": true|false, \"reason\": \"\u4E00\u53E5\u8BDD\"}\u3002";
+/** Memory rubric: accurate, usable, reusable, non-duplicate, fresh, right-home. */
+export declare const REVIEW_MEMORY_SYSTEM_PROMPT = "\u4F60\u662F\u8BB0\u5FC6\u5E93\u7684\u5185\u5BB9\u8BC4\u5BA1\u3002\u5148\u8FC7\u5F52\u5C5E\u4E09\u6D4B\uFF08\u4E3B\u8BED\u662F\u7528\u6237/\u73AF\u5883\u8FD8\u662F\u67D0\u4ED3\u5E93\u7684\u67D0\u7248\u673A\u5236\uFF1B\u62B9\u6389\u4ED3\u5E93\u540D\u7248\u672C\u53F7\u673A\u5236\u540D\u540E\u8FD8\u6210\u7ACB\u5417\uFF1B\u4ED3\u5E93\u4E8B\u5B9E\u4F4F ADR/docs\uFF09\uFF1A\u5F52\u5C5E\u4E0D\u8FC7\u76F4\u63A5 quarantine \u5E76\u6CE8\u660E\u653E\u9519\u5BB6\u3002\u518D\u6309\u4E94\u7EF4\u5224\uFF1A\u2460\u51C6\u786E\uFF08\u4E8B\u5B9E\u53EF\u4FE1\u3001\u65E0\u5E7B\u89C9\uFF09\uFF1B\u2461\u53EF\u7528\uFF08Why/How \u5177\u4F53\u3001\u53EF\u6267\u884C\uFF0C\u53CD\u9988\u7C7B\u5FC5\u987B\u5E26 Why \u4E0E How to apply\uFF09\uFF1B\u2462\u53EF\u590D\u7528\uFF08\u672A\u6765\u4F1A\u8BDD\u7528\u5F97\u4E0A\uFF09\uFF1B\u2463\u4E0D\u91CD\u590D\uFF08\u4E0E\u5019\u9009\u65E7\u6761\u6BD4\uFF0C\u6709\u91CD\u590D\u5C31\u6307\u51FA\u6765\uFF09\uFF1B\u2464\u4E0D\u8FC7\u65F6\uFF08\u672A\u88AB\u7EA0\u6B63\u94FE\u53D6\u4EE3\uFF09\u3002\u53EA\u8F93\u51FA JSON\uFF1A{\"verdict\": \"keep|rewrite|quarantine\", \"reason\": \"\u4E00\u53E5\u8BDD\", \"title\"?: \"\u6539\u5199\u540E\u6807\u9898\", \"description\"?: \"\u6539\u5199\u540E\u94A9\u5B50\", \"body\"?: \"\u6539\u5199\u540E\u6B63\u6587\u5168\u6587\", \"corrects\"?: \"\u88AB\u53D6\u4EE3\u7684\u540C\u533A\u6761\u76EEid\"}\u3002verdict \u8BF4\u660E\uFF1Akeep=\u4E0D\u52A8\uFF1Brewrite=\u7ED9\u51FA\u6539\u5199\u540E\u5168\u6587\u4E09\u4EF6\uFF1Bquarantine=\u6D89\u5BC6/\u6709\u5BB3/\u5B8C\u5168\u8FC7\u65F6/\u653E\u9519\u5BB6\u3002\u7981\u5220\u9664\uFF0C\u7981\u7F16\u9020\u5F15\u7528\u3002";
+/** Skill rubric: trigger precision, executability, reusability, non-duplicate, right-home. */
+export declare const REVIEW_SKILL_SYSTEM_PROMPT = "\u4F60\u662F\u6280\u80FD\u5E93\u7684\u5185\u5BB9\u8BC4\u5BA1\u3002\u5148\u8FC7\u5F52\u5C5E\u4E09\u6D4B\uFF08\u6280\u80FD\u662F\u5426\u7ED1\u5B9A\u5355\u4E00\u4ED3\u5E93\u7684\u8DEF\u5F84\u4E0E\u673A\u5236\uFF1B\u62B9\u6389\u4ED3\u5E93\u540D\u540E\u8FD8\u6210\u7ACB\u5417\uFF09\uFF1A\u653E\u9519\u5BB6\u76F4\u63A5 quarantine\u3002\u518D\u6309\u56DB\u7EF4\u5224\uFF1A\u2460\u89E6\u53D1\u7CBE\u5EA6\uFF08description \u662F\u5426\u8BA9\u4EBA/\u6A21\u578B\u5728\u8BE5\u7528\u65F6\u60F3\u8D77\u3001\u4E0D\u8BE5\u7528\u65F6\u4E0D\u6253\u6270\uFF09\uFF1B\u2461\u6B63\u6587\u53EF\u6267\u884C\u5B8C\u5907\uFF08\u6B65\u9AA4\u5B8C\u6574\u3001\u8F93\u5165\u8F93\u51FA\u660E\u786E\uFF09\uFF1B\u2462\u53EF\u590D\u7528\uFF08\u8DE8\u4EFB\u52A1\u53EF\u7528\uFF09\uFF1B\u2463\u4E0D\u91CD\u590D\uFF08\u4E0E\u5019\u9009\u65E2\u6709\u6280\u80FD\u6BD4\uFF09\u3002\u53EA\u8F93\u51FA JSON\uFF1A{\"verdict\": \"keep|rewrite|quarantine\", \"reason\": \"\u4E00\u53E5\u8BDD\", \"title\"?: \"\u6539\u5199\u540E\u6807\u9898\", \"description\"?: \"\u6539\u5199\u540E\u94A9\u5B50\", \"body\"?: \"\u6539\u5199\u540E\u6B63\u6587\u5168\u6587\"}\u3002verdict \u8BF4\u660E\uFF1Akeep=\u4E0D\u52A8\uFF1Brewrite=\u7ED9\u51FA\u6539\u5199\u540E\u5168\u6587\u4E09\u4EF6\uFF1Bquarantine=\u6D89\u5BC6/\u6709\u5BB3/\u5B8C\u5168\u8FC7\u65F6/\u653E\u9519\u5BB6\u3002\u7981\u5220\u9664\u3002";
+export interface GateVerdict {
+    worthReview: boolean;
+    reason: string;
+}
+export interface EvolveVerdict {
+    verdict: "keep" | "rewrite" | "quarantine";
+    reason: string;
+    title?: string | undefined;
+    description?: string | undefined;
+    body?: string | undefined;
+    corrects?: string | undefined;
+}
+/** The cheap-gate prompt for one item (id + hook + body, truncated by caller). */
+export declare function gatePrompt(entry: {
+    kind: string;
+    id: string;
+    title: string;
+    description: string;
+    body: string;
+}): string;
+/** The evolve prompt for one item plus its similar neighbors for dup-checking. */
+export declare function evolvePrompt(entry: {
+    id: string;
+    title: string;
+    description: string;
+    body: string;
+    extra?: string | undefined;
+}, neighbors: readonly {
+    id: string;
+    title: string;
+    description: string;
+}[]): string;
+/** Parse the cheap-gate answer; `null` when the model did not return verdict JSON. */
+export declare function parseGateAnswer(text: string): GateVerdict | null;
+/** Parse the evolve answer; `null` when the model did not return verdict JSON. */
+export declare function parseEvolveAnswer(text: string): EvolveVerdict | null;
+//# sourceMappingURL=review-prompt.d.ts.map

package/lib/review-prompt.js

@@ -0,0 +1,105 @@
+/**
+ * The content-quality review prompts (the model side of the autonomous
+ * evolution loop): a cheap gate that decides whether an item deserves a
+ * close look, and the evolve pass that judges content quality dimension by
+ * dimension and proposes the rewrite. Code owns every consequence — the
+ * model only returns verdict JSON, never touches the store.
+ *
+ * Memories are judged on five dimensions (accurate, usable, reusable,
+ * non-duplicate, fresh); skills on four (description trigger precision,
+ * body executability, reusability, non-duplicate). The prompts are
+ * deliberately behavior-blind: there are no task datasets here (unlike
+ * Hermes GEPA), only the item text plus its provenance.
+ */
+export const REVIEW_CHEAP_SYSTEM_PROMPT = "你是记忆库的内容评审门。先过归属三测：①主语是用户/环境还是某仓库的某版机制（版本号、阈值、开关名出现即后者）；②抹掉仓库名版本号机制名后句子还成立吗；③仓库事实住 ADR/docs，记忆只收跟人走的。归属不过直接判 false 并注明放错家。再判值不值得细看：有明确事实、做法、偏好或可复用规程就批；空话、纯过程记录、一次性的瞬时输出、当轮已可推导的一律拒。只输出 JSON：{\"worthReview\": true|false, \"reason\": \"一句话\"}。";
+/** Memory rubric: accurate, usable, reusable, non-duplicate, fresh, right-home. */
+export const REVIEW_MEMORY_SYSTEM_PROMPT = "你是记忆库的内容评审。先过归属三测（主语是用户/环境还是某仓库的某版机制；抹掉仓库名版本号机制名后还成立吗；仓库事实住 ADR/docs）：归属不过直接 quarantine 并注明放错家。再按五维判：①准确（事实可信、无幻觉）；②可用（Why/How 具体、可执行，反馈类必须带 Why 与 How to apply）；③可复用（未来会话用得上）；④不重复（与候选旧条比，有重复就指出来）；⑤不过时（未被纠正链取代）。只输出 JSON：{\"verdict\": \"keep|rewrite|quarantine\", \"reason\": \"一句话\", \"title\"?: \"改写后标题\", \"description\"?: \"改写后钩子\", \"body\"?: \"改写后正文全文\", \"corrects\"?: \"被取代的同区条目id\"}。verdict 说明：keep=不动；rewrite=给出改写后全文三件；quarantine=涉密/有害/完全过时/放错家。禁删除，禁编造引用。";
+/** Skill rubric: trigger precision, executability, reusability, non-duplicate, right-home. */
+export const REVIEW_SKILL_SYSTEM_PROMPT = "你是技能库的内容评审。先过归属三测（技能是否绑定单一仓库的路径与机制；抹掉仓库名后还成立吗）：放错家直接 quarantine。再按四维判：①触发精度（description 是否让人/模型在该用时想起、不该用时不打扰）；②正文可执行完备（步骤完整、输入输出明确）；③可复用（跨任务可用）；④不重复（与候选既有技能比）。只输出 JSON：{\"verdict\": \"keep|rewrite|quarantine\", \"reason\": \"一句话\", \"title\"?: \"改写后标题\", \"description\"?: \"改写后钩子\", \"body\"?: \"改写后正文全文\"}。verdict 说明：keep=不动；rewrite=给出改写后全文三件；quarantine=涉密/有害/完全过时/放错家。禁删除。";
+/** The cheap-gate prompt for one item (id + hook + body, truncated by caller). */
+export function gatePrompt(entry) {
+    return [
+        `kind=${entry.kind} id=${entry.id}`,
+        `title=${entry.title}`,
+        `钩子=${entry.description}`,
+        "全文：",
+        entry.body,
+    ].join("\n");
+}
+/** The evolve prompt for one item plus its similar neighbors for dup-checking. */
+export function evolvePrompt(entry, neighbors) {
+    const lines = [
+        `id=${entry.id}`,
+        `title=${entry.title}`,
+        `钩子=${entry.description}`,
+        "全文：",
+        entry.body,
+    ];
+    if (entry.extra !== undefined && entry.extra.length > 0) {
+        lines.push("", entry.extra);
+    }
+    if (neighbors.length > 0) {
+        lines.push("", "同类候选（判重复用）：");
+        for (const neighbor of neighbors) {
+            lines.push(`- id=${neighbor.id} title=${neighbor.title} 钩子=${neighbor.description}`);
+        }
+    }
+    return lines.join("\n");
+}
+/** Parse the cheap-gate answer; `null` when the model did not return verdict JSON. */
+export function parseGateAnswer(text) {
+    const parsed = parseJsonObject(text);
+    if (parsed === null || typeof parsed.worthReview !== "boolean") {
+        return null;
+    }
+    return {
+        worthReview: parsed.worthReview,
+        reason: typeof parsed.reason === "string" && parsed.reason.length > 0 ? parsed.reason.slice(0, 300) : "无理由",
+    };
+}
+/** Parse the evolve answer; `null` when the model did not return verdict JSON. */
+export function parseEvolveAnswer(text) {
+    const parsed = parseJsonObject(text);
+    if (parsed === null || (parsed.verdict !== "keep" && parsed.verdict !== "rewrite" && parsed.verdict !== "quarantine")) {
+        return null;
+    }
+    const verdict = {
+        verdict: parsed.verdict,
+        reason: typeof parsed.reason === "string" && parsed.reason.length > 0 ? parsed.reason.slice(0, 300) : "无理由",
+    };
+    if (typeof parsed.title === "string" && parsed.title.length > 0) {
+        verdict.title = parsed.title.slice(0, 500);
+    }
+    if (typeof parsed.description === "string" && parsed.description.length > 0) {
+        verdict.description = parsed.description.slice(0, 1000);
+    }
+    if (typeof parsed.body === "string" && parsed.body.length > 0) {
+        verdict.body = parsed.body.slice(0, 65536);
+    }
+    if (typeof parsed.corrects === "string" && parsed.corrects.length > 0) {
+        verdict.corrects = parsed.corrects.slice(0, 200);
+    }
+    return verdict;
+}
+/** Extract the first JSON object from model output (fenced or bare). */
+function parseJsonObject(text) {
+    const fenced = text.match(/```(?:json)?\s*([\s\S]*?)```/);
+    const candidate = (fenced?.[1] ?? text).trim();
+    const start = candidate.indexOf("{");
+    const end = candidate.lastIndexOf("}");
+    if (start === -1 || end <= start) {
+        return null;
+    }
+    try {
+        const parsed = JSON.parse(candidate.slice(start, end + 1));
+        /* v8 ignore next 3 — the slice always starts with `{`, so a
+        successful parse is an object by construction; the guard stays. */
+        return typeof parsed === "object" && parsed !== null && !Array.isArray(parsed)
+            ? parsed
+            : null;
+    }
+    catch {
+        return null;
+    }
+}
+//# sourceMappingURL=review-prompt.js.map

package/lib/review.d.ts

@@ -0,0 +1,99 @@
+import type { MemoryStore } from "./store.js";
+/** The minimal LLM face review calls (same duck-typed host service as extraction). */
+export interface ReviewLlm {
+    stream(options: Record<string, unknown>): AsyncIterable<unknown>;
+}
+export interface ReviewDeps {
+    store: MemoryStore;
+    llm: ReviewLlm | undefined;
+}
+export interface ReviewCandidate {
+    workspaceId: string;
+    kind: "memory" | "skill";
+    id: string;
+    /** 0 = new, 1 = corrects-linked requeue, 2 = stale requeue. */
+    priority: number;
+    title: string;
+    description: string;
+    body: string;
+    /** Memory type, present only for memory candidates. */
+    memoryType?: string | undefined;
+    /** Skill scope, present only for skill candidates. */
+    scope?: "project" | "global" | undefined;
+}
+export interface ReviewRunOptions {
+    /** UTC day `YYYY-MM-DD` the hard cap accounts against. */
+    day: string;
+    /** Hard cap for the day (100). Candidates past it count as dropped. */
+    budget: number;
+    /** Trial mode: verdicts ledgered, nothing applied or quarantined. */
+    dryRun: boolean;
+    /** Max items per run (batching keeps one prompt small). */
+    batchSize?: number | undefined;
+    /** Days of untouched age before an item requeues (default 30). */
+    staleDays?: number | undefined;
+    provider?: string | undefined;
+    model?: string | undefined;
+    sessionId?: string | undefined;
+    runId?: string | undefined;
+}
+export interface ReviewRunSummary {
+    reviewed: number;
+    approved: number;
+    applied: number;
+    dropped: number;
+    backlog: number;
+    cheapToken: number;
+    evolveToken: number;
+}
+/** A rewrite lands only when the body-length ratio stays within this bound. */
+export declare const REWRITE_MAGNITUDE_CAP = 0.3;
+/** Relative body change: |new-old| / max(old, 1). Above the cap → quarantine. */
+export declare function rewriteMagnitude(oldBody: string, newBody: string): number;
+/**
+ * Collect review candidates across every workspace carrying active items:
+ * new items first (newest first), then corrects-linked requeues, then stale
+ * requeues (30 days untouched). bounded by `limit`.
+ */
+export declare function collectCandidates(store: MemoryStore, opts?: {
+    now?: number | undefined;
+    staleDays?: number | undefined;
+    limit?: number | undefined;
+}): ReviewCandidate[];
+/**
+ * Run one review batch: cheap gate per item, evolve pass on approval,
+ * safeguards before landing, one `item-review` row per terminal path, one
+ * daily merge at the end. Never throws: LLM and parse failures degrade to
+ * skip rows (the run keeps going), exactly like extraction's fail-ledgered
+ * discipline.
+ */
+export declare function runReviewBatch(deps: ReviewDeps, candidates: readonly ReviewCandidate[], opts: ReviewRunOptions): Promise<ReviewRunSummary>;
+/** One-line error head for ledger rows (fail-loud shape, capped length). */
+export declare function failureHead(error: unknown, max?: number): string;
+/** Current UTC day `YYYY-MM-DD` for the hard-cap account. */
+export declare function todayUtc(): string;
+/** The event wiring surface (duck-typed cordis context, mirrors extraction). */
+export interface ReviewSchedulerHost {
+    on(event: string, handler: (...args: never[]) => void): unknown;
+    logger(name: string): {
+        info(message: string): void;
+        warn(message: string): void;
+    };
+}
+export interface ReviewSchedulerConfig {
+    enabled: boolean;
+    dryRun: boolean;
+    dailyBudget: number;
+    batchSize: number;
+    staleDays: number;
+    /** Minimum ms between two drains (default 5 minutes). */
+    minIntervalMs: number;
+}
+/**
+ * Wire the review triggers onto the host context. Global single-flight with
+ * a minimum interval: turn boundaries arm the pending workspace, compaction
+ * and session close drain immediately. Each drain reviews one batch bounded
+ * by the remaining daily budget — never more.
+ */
+export declare function createReviewScheduler(host: ReviewSchedulerHost, deps: ReviewDeps, config: ReviewSchedulerConfig): () => void;
+//# sourceMappingURL=review.d.ts.map