@lmzhen/dsh-evolution-review 0.3.16 → 0.3.18

package/README.md

@@ -24,6 +24,8 @@ Independent of request-prefix construction. This package does not alter the asse
 - Review subagents inherit the host preset; Anchored Standard deployments rely on the default `skill_search`/`skill_load` allow-list.
 - Review subagents run as `spawn` children on the deployment default preset rather than inheriting the parent agent's composition (`fork`): a fork child is always promoted by the Anchored Standard bootstrap and its narrowed resident catalog would drop the plain `skill` tool from the review allow-list.
 - The review request text is redacted for credential-shaped patterns before it reaches the subagent, but redaction is pattern-based and best-effort, not a security boundary.
+- Read-before-write tracks only reads through the `skill` tool. `skill_manage` has no per-skill read action (`list`/`review` are whole-library, not targeted at one name), so a skill that was only listed via `skill_manage` is not marked as read — a background review may still reject a patch to it until it is actually loaded.
+- The completion-channel counters (`cumulativeToolCalls` / `completionInjected`) are in-memory only. A process restart resets them, which is accepted behavior: the completion review is a one-per-session post-task adaptation and a restart is treated as a fresh conversation boundary. The cadence state (`turnsSinceMemory` / `turnsSinceSkill`) is persisted via `ReviewState` and survives restart.
 
 ## Configuration
 

package/lib/index.js

@@ -9,10 +9,10 @@ import { validateEvolutionPlan } from "@lmzhen/dsh-evolution-plan-validator";
 * @module @lmzhen/dsh-evolution-review
 */
 const name = "evolution-review";
-const inject = ["agents", "tools"];
+const inject = ["agents"];
 const Config = z.object({
 	reviewEnabled: z.boolean().default(true),
-	reviewMode: z.string().default("subagent"),
+	reviewMode: z.union([z.const("subagent"), z.const("inject")]).default("subagent"),
 	memoryInterval: z.number().default(DEFAULT_REVIEW_MEMORY_INTERVAL),
 	skillInterval: z.number().default(DEFAULT_REVIEW_SKILL_INTERVAL),
 	reviewToolAllow: z.array(z.string()).default(["skill"]),
@@ -31,6 +31,7 @@ function apply(ctx, rawConfig) {
 	const turnStarts = /* @__PURE__ */ new Map();
 	const cumulativeToolCalls = /* @__PURE__ */ new Map();
 	const completionInjected = /* @__PURE__ */ new Set();
+	let reviewInFlight = false;
 	const policy = () => ctx.get("evolutionPolicy")?.get();
 	ctx.on("session/event", (session, event) => {
 		if (event.type === "turn/start") turnStarts.set(session.id, session.seq - 1);
@@ -44,6 +45,14 @@ function apply(ctx, rawConfig) {
 		onTurnEnd(session, event);
 	});
 	async function onTurnEnd(session, event) {
+		try {
+			await runOnTurnEnd(session, event);
+		} catch (error) {
+			ctx.logger.warn(`dsh-evolution-review: turn-end review pipeline failed: ${error instanceof Error ? error.message : String(error)}`);
+			ctx.emit("evolution/review-error", { sessionId: session.id });
+		}
+	}
+	async function runOnTurnEnd(session, event) {
 		if (!config.reviewEnabled) return;
 		if (session.header.origin === "subagent") return;
 		const agent = ctx.agents.get(session.id);
@@ -68,14 +77,14 @@ function apply(ctx, rawConfig) {
 		const cumulative = (cumulativeToolCalls.get(session.id) ?? 0) + signal.toolCalls;
 		cumulativeToolCalls.set(session.id, cumulative);
 		if (kind) {
-			ctx.emit("evolution/review-scheduled", {
+			if (await trySubagentReview(session, agent, kind, signal)) ctx.emit("evolution/review-scheduled", {
 				sessionId: session.id,
 				kind,
 				toolCalls: signal.toolCalls,
 				userChars: signal.userChars,
 				assistantChars: signal.assistantChars
 			});
-			if (!await trySubagentReview(session, agent, kind, signal)) agent.inject(createUserMessage({
+			else agent.inject(createUserMessage({
 				content: [{
 					type: "text",
 					text: reviewPrompt(kind)
@@ -118,9 +127,11 @@ function apply(ctx, rawConfig) {
 		if ((policy()?.reviewMode ?? config.reviewMode) === "inject") return false;
 		const subagents = ctx.get("subagents");
 		if (!subagents) return false;
+		if (reviewInFlight) return false;
+		reviewInFlight = true;
 		try {
-			const routingPolicy = ctx.get("evolutionPolicy");
-			const model = kind === "memory" ? routingPolicy?.get().memoryReviewModel ?? "deepseek-v4-flash" : routingPolicy?.get().skillReviewModel ?? "deepseek-v4-pro";
+			const snapshot = policy();
+			const model = kind === "memory" ? snapshot?.memoryReviewModel ?? "deepseek-v4-flash" : snapshot?.skillReviewModel ?? "deepseek-v4-pro";
 			const reviewText = redactSecrets(buildReviewRequest(session, kind, signal, config.reviewContextMessages, config.reviewMessageChars));
 			const agentOptions = { model };
 			if (config.reviewProvider) agentOptions.provider = config.reviewProvider;
@@ -154,9 +165,12 @@ function apply(ctx, rawConfig) {
 			});
 			try {
 				const result = await run.result;
-				if (!result.structured) return true;
+				if (!result.structured) {
+					ctx.emit("evolution/review-error", { sessionId: session.id });
+					ctx.logger.warn("dsh-evolution-review: review subagent returned no structured plan");
+					return false;
+				}
 				const childReads = run.localAgent ? collectReadSkillNames(run.localAgent.session) : /* @__PURE__ */ new Set();
-				const snapshot = policy();
 				const plan = result.structured;
 				const policyFingerprint = fingerprintPolicy(snapshot);
 				const validation = validateEvolutionPlan(plan, {
@@ -169,7 +183,8 @@ function apply(ctx, rawConfig) {
 				});
 				const acceptedSkillOps = validation.accepted.skillOps ?? [];
 				const skippedUnread = filterUnreadSkillOps(acceptedSkillOps, new Set([...collectReadSkillNames(session), ...childReads]));
-				const actions = await executePlan(validation.accepted);
+				const executed = await executePlan(validation.accepted, session.id);
+				const actions = executed.actions;
 				const evidenceQuotes = [...validation.accepted.memoryOps ?? [], ...acceptedSkillOps].reduce((total, op) => total + (Array.isArray(op.evidence) ? op.evidence.length : 0), 0);
 				ctx.emit("evolution/plan-applied", {
 					sessionId: session.id,
@@ -181,18 +196,22 @@ function apply(ctx, rawConfig) {
 					evidenceQuotes,
 					estimatedInputChars: reviewText.length
 				});
-				if (actions.length > 0) agent.inject(createUserMessage({
-					content: [{
-						type: "text",
-						text: `💾 Self-improvement review: ${actions.join(" · ")}`
-					}],
-					source: {
-						kind: "plugin",
-						plugin: "dsh-evolution-review",
-						form: "notice",
-						summary: "self-improvement review"
-					}
-				}));
+				if (actions.length > 0) {
+					const applied = actions.join(" · ");
+					const note = executed.ok ? "" : "\n部分操作失败。以下操作已应用，请勿重复执行。";
+					agent.inject(createUserMessage({
+						content: [{
+							type: "text",
+							text: `💾 Self-improvement review: ${applied}${note}`
+						}],
+						source: {
+							kind: "plugin",
+							plugin: "dsh-evolution-review",
+							form: "notice",
+							summary: "self-improvement review"
+						}
+					}));
+				}
 				return true;
 			} finally {
 				try {
@@ -204,13 +223,16 @@ function apply(ctx, rawConfig) {
 		} catch (error) {
 			ctx.logger.warn(`dsh-evolution-review: subagent review failed: ${error instanceof Error ? error.message : String(error)}`);
 			return false;
+		} finally {
+			reviewInFlight = false;
 		}
 	}
-	async function executePlan(plan) {
+	async function executePlan(plan, sessionId) {
 		const memory = ctx.get("memory");
 		const approval = ctx.get("evolutionApproval");
 		const origins = resolveOrigins(void 0, true);
 		const actions = [];
+		let ok = true;
 		for (const op of plan.memoryOps ?? []) {
 			if (!Array.isArray(op.evidence) || op.evidence.length === 0) continue;
 			const normalized = {
@@ -220,6 +242,7 @@ function apply(ctx, rawConfig) {
 				old_text: op.old_text
 			};
 			if ((approval ? await runApproved("memory", `memory ${normalized.target} ${normalized.action}`, normalized, normalized) : await memory?.applyBatch(normalized.target, [normalized]))?.ok) actions.push("Memory updated");
+			else ok = false;
 		}
 		for (const op of plan.skillOps ?? []) {
 			if (!Array.isArray(op.evidence) || op.evidence.length === 0 || !op.name) continue;
@@ -232,8 +255,12 @@ function apply(ctx, rawConfig) {
 				origin: origins.library
 			};
 			if ((approval ? await runApproved("skill", `skill ${op.action ?? "patch"} ${op.name}`, runnerArgs, runnerArgs) : await executeSkillDirect(args))?.ok) actions.push(`Skill ${op.name} ${op.action ?? "patch"}`);
+			else ok = false;
 		}
-		return actions;
+		return {
+			actions,
+			ok
+		};
 		async function runApproved(kind, summary, stored, runnerArgs) {
 			if (!approval) return void 0;
 			if (approval.isEnabled === true && !approval.hasRunner(kind)) {
@@ -247,14 +274,15 @@ function apply(ctx, rawConfig) {
 				kind,
 				summary,
 				args: stored,
-				origin: origins.approval
+				origin: origins.approval,
+				...sessionId ? { sessionId } : {}
 			});
 			if (decision.action === "staged") return {
 				ok: false,
 				message: decision.message
 			};
 			if (approval.isEnabled === false) return await runnerDirect(kind, runnerArgs);
-			return await approval.run(kind, runnerArgs);
+			return await approval.run(kind, runnerArgs, { interface: "background_review" });
 		}
 		/** Direct execution for the approval-disabled case (parallel to executeSkillDirect). */
 		async function runnerDirect(kind, args) {
@@ -333,12 +361,16 @@ function apply(ctx, rawConfig) {
 function shouldCompletionReview(reason, sessionToolCalls, minToolCalls) {
 	return reason?.kind === "completed" && sessionToolCalls >= minToolCalls;
 }
-/** Skill names this session loaded (read-before-write source for the background review). */
+/** Skill names this session loaded (read-before-write source for the background review).
+* Only the real `skill` tool is a read (E-59e): the platform has no `skill_load`/
+* `skill_search` discovery pair, so that branch was dead. `skill_manage` has no
+* per-skill read action (its `list`/`review` are whole-library), so a specific
+* skill read through it cannot be tracked — see README Known Limitations. */
 function collectReadSkillNames(session) {
 	const names = /* @__PURE__ */ new Set();
 	for (const event of session.events) {
 		if (event.type !== "tool/call") continue;
-		if (event.data.name !== "skill" && event.data.name !== "skill_load") continue;
+		if (event.data.name !== "skill") continue;
 		const raw = event.data.arguments;
 		let parsed = {};
 		if (typeof raw === "string") try {

package/lib/types/index.d.ts

@@ -8,7 +8,7 @@ export declare const name = "evolution-review";
 export declare const inject: string[];
 export interface Config {
     reviewEnabled?: boolean;
-    reviewMode?: string;
+    reviewMode?: 'subagent' | 'inject';
     memoryInterval?: number;
     skillInterval?: number;
     /**

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@lmzhen/dsh-evolution-review",
   "description": "Background review orchestration (community build)",
-  "version": "0.3.16",
+  "version": "0.3.18",
   "publishConfig": {
     "access": "public"
   },
@@ -33,8 +33,8 @@
   "license": "MIT",
   "dependencies": {
     "@deepseek-ai/schemastery": "^3.18.1",
-    "@lmzhen/dsh-evolution-core": "^0.3.16",
-    "@lmzhen/dsh-evolution-plan-validator": "^0.3.16"
+    "@lmzhen/dsh-evolution-core": "^0.3.18",
+    "@lmzhen/dsh-evolution-plan-validator": "^0.3.18"
   },
   "peerDependencies": {
     "@deepseek-ai/cordis": "^4.0.1",
@@ -43,7 +43,8 @@
     "@deepseek-ai/dsh-llm": "^0.1.1-rc.2",
     "@deepseek-ai/dsh-session": "^0.1.1-rc.2",
     "@deepseek-ai/dsh-tools": "^0.1.1-rc.2",
-    "@lmzhen/dsh-evolution-state": "^0.3.16"
+    "@lmzhen/dsh-evolution-state": "^0.3.18",
+    "@lmzhen/dsh-evolution-policy": "^0.3.18"
   },
   "devDependencies": {
     "@deepseek-ai/dsh-agent": "^0.1.1-rc.2",
@@ -54,9 +55,9 @@
     "@deepseek-ai/dsh-session-persistence": "^0.1.1-rc.2",
     "@deepseek-ai/dsh-session-persistence-jsonl": "^0.1.1-rc.2",
     "@deepseek-ai/dsh-tools": "^0.1.1-rc.2",
-    "@lmzhen/dsh-evolution-approval": "^0.3.16",
-    "@lmzhen/dsh-evolution-core": "^0.3.16",
-    "@lmzhen/dsh-evolution-plan-validator": "^0.3.16",
-    "@lmzhen/dsh-evolution-state": "^0.3.16"
+    "@lmzhen/dsh-evolution-approval": "^0.3.18",
+    "@lmzhen/dsh-evolution-core": "^0.3.18",
+    "@lmzhen/dsh-evolution-plan-validator": "^0.3.18",
+    "@lmzhen/dsh-evolution-state": "^0.3.18"
   }
 }