@lmzhen/dsh-evolution-review 0.1.0-rc.9 → 0.2.0-rc.1

package/lib/index.js

@@ -1,7 +1,7 @@
 import { createHash, randomUUID } from "node:crypto";
 import z from "@deepseek-ai/schemastery";
-import { CallId, createUserMessage } from "@deepseek-ai/dsh-llm";
-import { PROMPT_BUNDLE, advanceReview, foldTurn, reviewPrompt, verifyPromptBundle } from "@lmzhen/dsh-evolution-core";
+import { createUserMessage } from "@deepseek-ai/dsh-llm";
+import { COMPLETION_SKILL_REVIEW_PROMPT, DEFAULT_MAX_OPS_PER_PLAN, DEFAULT_MEMORY_CHAR_LIMIT, DEFAULT_REVIEW_MEMORY_INTERVAL, DEFAULT_REVIEW_SKILL_INTERVAL, DEFAULT_SKILL_CONTENT_CHARS, DEFAULT_SKILL_REVIEW_COMPLETION_MIN_TOOL_CALLS, DEFAULT_SKILL_REVIEW_TRIGGER, DEFAULT_USER_CHAR_LIMIT, PROMPT_BUNDLE, SkillLibrary, advanceReview, evolutionIoAdapter, foldTurn, resolveOrigins, reviewPrompt, verifyPromptBundle } from "@lmzhen/dsh-evolution-core";
 import { validateEvolutionPlan } from "@lmzhen/dsh-evolution-plan-validator";
 //#region lib/types/redact.js
 /**
@@ -42,28 +42,34 @@ const inject = ["agents", "tools"];
 const Config = z.object({
 	reviewEnabled: z.boolean().default(true),
 	reviewMode: z.string().default("subagent"),
-	memoryInterval: z.number().default(10),
-	skillInterval: z.number().default(10),
-	reviewToolAllow: z.array(z.string()).default([
-		"skill",
-		"skill_search",
-		"skill_load"
-	]),
+	memoryInterval: z.number().default(DEFAULT_REVIEW_MEMORY_INTERVAL),
+	skillInterval: z.number().default(DEFAULT_REVIEW_SKILL_INTERVAL),
+	reviewToolAllow: z.array(z.string()).default(["skill"]),
 	reviewTimeoutMs: z.number().default(12e4),
 	executionTimeoutMs: z.number().default(3e4),
 	reviewContextMessages: z.number().default(60),
 	reviewMessageChars: z.number().default(2e3),
 	reviewMaxDepth: z.number().default(0),
-	reviewProvider: z.string()
+	reviewProvider: z.string(),
+	skillReviewTrigger: z.string().default(DEFAULT_SKILL_REVIEW_TRIGGER),
+	skillReviewCompletionMinToolCalls: z.number().default(DEFAULT_SKILL_REVIEW_COMPLETION_MIN_TOOL_CALLS)
 });
 function apply(ctx, rawConfig) {
 	if (!verifyPromptBundle(PROMPT_BUNDLE)) throw new Error("dsh-evolution prompt bundle integrity check failed; refusing to schedule review work");
 	const config = rawConfig;
 	const turnStarts = /* @__PURE__ */ new Map();
+	const cumulativeToolCalls = /* @__PURE__ */ new Map();
+	const completionInjected = /* @__PURE__ */ new Set();
 	const policy = () => ctx.get("evolutionPolicy")?.get();
 	ctx.on("session/event", (session, event) => {
 		if (event.type === "turn/start") turnStarts.set(session.id, session.seq - 1);
 		if (event.type !== "turn/end") return;
+		if (turnStarts.size >= COUNTER_SWEEP_THRESHOLD || cumulativeToolCalls.size >= COUNTER_SWEEP_THRESHOLD || completionInjected.size >= COUNTER_SWEEP_THRESHOLD) {
+			const isAlive = (id) => ctx.agents.get(id) !== void 0;
+			sweepDeadSessionEntries(turnStarts, isAlive);
+			sweepDeadSessionEntries(cumulativeToolCalls, isAlive);
+			sweepDeadSessionEntries(completionInjected, isAlive);
+		}
 		onTurnEnd(session, event);
 	});
 	async function onTurnEnd(session, event) {
@@ -88,17 +94,52 @@ function apply(ctx, rawConfig) {
 			substantiveMinAgentChars: snapshot?.substantiveMinAgentChars ?? 500
 		});
 		await stateService?.saveReviewState(session.id, state);
-		if (!kind) return;
-		if (!await trySubagentReview(session, agent, kind, signal)) agent.inject(createUserMessage({
+		const cumulative = (cumulativeToolCalls.get(session.id) ?? 0) + signal.toolCalls;
+		cumulativeToolCalls.set(session.id, cumulative);
+		if (kind) {
+			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({
+				content: [{
+					type: "text",
+					text: reviewPrompt(kind)
+				}],
+				source: {
+					kind: "plugin",
+					plugin: "dsh-evolution-review",
+					form: "notice",
+					summary: "auto-review"
+				}
+			}));
+			return;
+		}
+		const trigger = config.skillReviewTrigger;
+		if (trigger !== "completion" && trigger !== "both") return;
+		if (completionInjected.has(session.id)) return;
+		if (!shouldCompletionReview(event.data.reason, cumulative, config.skillReviewCompletionMinToolCalls)) return;
+		completionInjected.add(session.id);
+		ctx.emit("evolution/review-scheduled", {
+			sessionId: session.id,
+			kind: "skill",
+			toolCalls: signal.toolCalls,
+			userChars: signal.userChars,
+			assistantChars: signal.assistantChars
+		});
+		agent.inject(createUserMessage({
 			content: [{
 				type: "text",
-				text: reviewPrompt(kind)
+				text: COMPLETION_SKILL_REVIEW_PROMPT
 			}],
 			source: {
 				kind: "plugin",
 				plugin: "dsh-evolution-review",
 				form: "notice",
-				summary: "auto-review"
+				summary: "completion review"
 			}
 		}));
 	}
@@ -122,7 +163,7 @@ function apply(ctx, rawConfig) {
 				signal: AbortSignal.timeout(config.reviewTimeoutMs),
 				maxDepth: config.reviewMaxDepth,
 				agentOptions,
-				persona: reviewPrompt(kind),
+				persona: reviewPrompt(kind, "plan"),
 				toolFilter: { allow: [...config.reviewToolAllow] },
 				outputSchema: {
 					type: "object",
@@ -140,51 +181,64 @@ function apply(ctx, rawConfig) {
 					}
 				}
 			});
-			const result = await run.result;
-			await run.dispose();
-			if (!result.structured) return true;
-			const snapshot = policy();
-			const plan = result.structured;
-			const policyFingerprint = fingerprintPolicy(snapshot);
-			const validation = validateEvolutionPlan(plan, {
-				sessionSeq: session.seq - 1,
-				maxOpsPerPlan: snapshot?.maxOpsPerPlan ?? 32,
-				protectedSkillNames: new Set(snapshot?.protectedSkillNames ?? []),
-				maxMemoryChars: snapshot?.memoryChars ?? 2200,
-				maxUserChars: snapshot?.userChars ?? 1375,
-				maxSkillContentChars: snapshot?.skillContentChars ?? 1e5
-			});
-			const actions = await executePlan(validation.accepted, agent);
-			const evidenceQuotes = [...validation.accepted.memoryOps ?? [], ...validation.accepted.skillOps ?? []].reduce((total, op) => total + (Array.isArray(op.evidence) ? op.evidence.length : 0), 0);
-			session.append("evolution/plan-applied", {
-				planId: randomUUID(),
-				policyFingerprint,
-				memoryApplied: actions.filter((action) => action.startsWith("Memory")).length,
-				skillApplied: actions.filter((action) => action.startsWith("Skill ")).length,
-				rejectedOps: validation.rejected.length,
-				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"
+			try {
+				const result = await run.result;
+				if (!result.structured) return true;
+				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, {
+					sessionSeq: session.seq - 1,
+					maxOpsPerPlan: snapshot?.maxOpsPerPlan ?? DEFAULT_MAX_OPS_PER_PLAN,
+					protectedSkillNames: new Set(snapshot?.protectedSkillNames ?? []),
+					maxMemoryChars: snapshot?.memoryChars ?? DEFAULT_MEMORY_CHAR_LIMIT,
+					maxUserChars: snapshot?.userChars ?? DEFAULT_USER_CHAR_LIMIT,
+					maxSkillContentChars: snapshot?.skillContentChars ?? DEFAULT_SKILL_CONTENT_CHARS
+				});
+				const acceptedSkillOps = validation.accepted.skillOps ?? [];
+				const skippedUnread = filterUnreadSkillOps(acceptedSkillOps, new Set([...collectReadSkillNames(session), ...childReads]));
+				const actions = await executePlan(validation.accepted);
+				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,
+					planId: randomUUID(),
+					policyFingerprint,
+					memoryApplied: actions.filter((action) => action.startsWith("Memory")).length,
+					skillApplied: actions.filter((action) => action.startsWith("Skill ")).length,
+					rejectedOps: validation.rejected.length + skippedUnread,
+					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"
+					}
+				}));
+				return true;
+			} finally {
+				try {
+					await run.dispose();
+				} catch (disposeError) {
+					ctx.logger.warn(`dsh-evolution-review: subagent dispose failed: ${disposeError instanceof Error ? disposeError.message : String(disposeError)}`);
 				}
-			}));
-			return true;
-		} catch {
+			}
+		} catch (error) {
+			ctx.logger.warn(`dsh-evolution-review: subagent review failed: ${error instanceof Error ? error.message : String(error)}`);
 			return false;
 		}
 	}
-	async function executePlan(plan, parent) {
+	async function executePlan(plan) {
 		const memory = ctx.get("memory");
 		const approval = ctx.get("evolutionApproval");
+		const origins = resolveOrigins(void 0, true);
 		const actions = [];
 		for (const op of plan.memoryOps ?? []) {
 			if (!Array.isArray(op.evidence) || op.evidence.length === 0) continue;
@@ -202,46 +256,177 @@ function apply(ctx, rawConfig) {
 				...op,
 				evidence: op.evidence
 			};
-			if ((approval ? await runApproved("skill", `skill ${op.action ?? "patch"} ${op.name}`, {
+			const runnerArgs = {
 				operation: args,
-				origin: "background_review"
-			}, args) : await executeSkillTool(parent, args))?.ok) actions.push(`Skill ${op.name} ${op.action ?? "patch"}`);
+				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} ${op.action ?? "patch"}`);
 		}
 		return actions;
 		async function runApproved(kind, summary, stored, runnerArgs) {
 			if (!approval) return void 0;
+			if (approval.isEnabled === true && !approval.hasRunner(kind)) {
+				ctx.logger.warn(`dsh-evolution-review: approval enabled but no replay runner registered for kind "${kind}" - skipping write (${summary})`);
+				return {
+					ok: false,
+					message: `Approval is enabled but no replay runner is registered for kind "${kind}"; write skipped (mount the tool that provides it, or disable approval).`
+				};
+			}
 			const decision = await approval.request({
 				kind,
 				summary,
 				args: stored,
-				origin: "background_review"
+				origin: origins.approval
 			});
 			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);
 		}
-		async function executeSkillTool(agent, args) {
-			return (await ctx.tools.execute({
-				callId: CallId(`evolution-${randomUUID()}`),
-				name: "skill_manage",
-				arguments: args,
-				agent,
-				signal: AbortSignal.timeout(config.executionTimeoutMs)
-			})).isError ? {
+		/** Direct execution for the approval-disabled case (parallel to executeSkillDirect). */
+		async function runnerDirect(kind, args) {
+			if (kind === "memory") {
+				const memory = ctx.get("memory");
+				const op = args;
+				if (!memory?.applyBatch) return void 0;
+				return await memory.applyBatch(op.target === "user" ? "user" : "memory", [{
+					action: op.action ?? "add",
+					facts: op.facts ?? op.content,
+					old_text: op.old_text
+				}]);
+			}
+			const wrapped = args ?? {};
+			if (!wrapped.operation) return void 0;
+			return await executeSkillDirect(wrapped.operation);
+		}
+		/**
+		* Approval-disabled path: execute the skill op through SkillLibrary with an
+		* EXPLICIT background_review origin. Going through ctx.tools.execute would
+		* make tool-skill-manage infer origin from the parent agent's header
+		* (not 'subagent'), silently escaping the .hermes-managed marker and the
+		* pinned write guard.
+		*/
+		async function executeSkillDirect(skillArgs) {
+			const io = ctx.get("evolutionIo");
+			if (!io) return {
 				ok: false,
-				message: "skill_manage execution failed"
-			} : {
-				ok: true,
-				message: "skill_manage executed"
+				message: "evolution-io service not mounted"
+			};
+			const library = new SkillLibrary(void 0, evolutionIoAdapter(() => io.provider()), void 0, (event) => {
+				ctx.emit("evolution/skill-mutated", event);
+			});
+			const op = skillArgs;
+			const name = op.name ?? "";
+			const origin = origins.library;
+			if (op.action === "create") {
+				const created = await library.create(name, op.content ?? "", origin);
+				if (created.ok) await ctx.get("skillUsage")?.markAgentCreated?.(name);
+				return created;
+			}
+			if (op.action === "edit" || op.action === "update") return await library.update(name, op.content ?? "", origin);
+			if (op.action === "patch") return await library.patch(name, op.old_string ?? "", op.new_string ?? "", op.file_path ?? "", false, origin);
+			if (op.action === "delete") {
+				const into = (op.absorbed_into ?? "").trim();
+				if (!into || !await library.read(into)) return {
+					ok: false,
+					message: "delete requires an existing absorbed_into target"
+				};
+				const archived = await library.archive(name, { absorbedInto: into });
+				if (archived.ok) await ctx.get("skillUsage")?.markArchived?.(name);
+				return archived;
+			}
+			if (op.action === "write_file") return await library.writeSupportFile(name, op.file_path ?? "", op.file_content ?? op.content ?? "", origin);
+			if (op.action === "remove_file") return await library.removeSupportFile(name, op.file_path ?? "", origin);
+			if (op.action === "restructure") {
+				const moves = (op.restructure ?? []).filter((move) => move !== null).map((move) => ({
+					heading: move.heading ?? "",
+					toFile: move.to_file ?? ""
+				}));
+				return await library.restructure(name, moves, origin);
+			}
+			return {
+				ok: false,
+				message: `Unknown skill action "${op.action ?? ""}"`
 			};
 		}
 	}
 	ctx.effect(() => () => {
 		turnStarts.clear();
+		cumulativeToolCalls.clear();
+		completionInjected.clear();
 	}, "dsh-evolution-review.cleanup");
 }
+/** Completion-channel decision: task finished normally AND the session is proven long. */
+function shouldCompletionReview(reason, sessionToolCalls, minToolCalls) {
+	return reason?.kind === "completed" && sessionToolCalls >= minToolCalls;
+}
+/** Skill names this session loaded (read-before-write source for the background review). */
+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;
+		const raw = event.data.arguments;
+		let parsed = {};
+		if (typeof raw === "string") try {
+			parsed = JSON.parse(raw);
+		} catch {
+			continue;
+		}
+		else parsed = raw ?? {};
+		const name = typeof parsed.name === "string" ? parsed.name : typeof parsed.skill === "string" ? parsed.skill : "";
+		if (name) names.add(name);
+	}
+	return names;
+}
+/** Map/set size that triggers a dead-session counter sweep (bounded, not a hard cap). */
+const COUNTER_SWEEP_THRESHOLD = 128;
+/**
+* Remove every entry whose session is no longer live (rc.42 audit P1-10):
+* `turnStarts` / `cumulativeToolCalls` / `completionInjected` are keyed by
+* SessionId with no platform session-end hook to prune against, so they grew
+* unbounded over a long-lived host. Works for maps and sets; returns the
+* number of removed entries.
+*/
+function sweepDeadSessionEntries(entries, isAlive) {
+	let removed = 0;
+	for (const id of [...entries.keys()]) if (!isAlive(id)) {
+		entries.delete(id);
+		removed += 1;
+	}
+	return removed;
+}
+/**
+* Drop mutating ops whose target was not read this session, in place.
+* Create is exempt (no read required to author a new skill). Covers the same
+* mutating surface Hermes guards (edit/patch/write_file/remove_file), so a
+* background review cannot blind-touch support files or edits of skills it
+* never loaded. Returns the count of dropped ops so the plan event can report
+* them as rejected.
+*/
+function filterUnreadSkillOps(ops, readNames) {
+	const READ_REQUIRED = [
+		"edit",
+		"update",
+		"patch",
+		"delete",
+		"write_file",
+		"remove_file",
+		"restructure"
+	];
+	let dropped = 0;
+	for (let index = ops.length - 1; index >= 0; index -= 1) {
+		const op = ops[index];
+		if (!op) continue;
+		if (op.action !== void 0 && READ_REQUIRED.includes(op.action) && op.name && !readNames.has(op.name)) {
+			ops.splice(index, 1);
+			dropped += 1;
+		}
+	}
+	return dropped;
+}
 function fingerprintPolicy(snapshot) {
 	try {
 		return createHash("sha256").update(JSON.stringify(snapshot)).digest("hex").slice(0, 12);
@@ -256,13 +441,31 @@ function buildReviewRequest(session, kind, signal, maxMessages, maxMessageChars)
 		const text = message.content.map((block) => block.type === "text" ? block.text : "").join(" ").trim();
 		if (text) messages.push(`${message.role.toUpperCase()}: ${text.slice(0, maxMessageChars)}`);
 	}
+	const toolLines = [];
+	const events = session.events;
+	for (let index = events.length - 1; index >= 0 && toolLines.length < 12; index -= 1) {
+		const event = events[index];
+		if (event?.type === "tool/call") {
+			const data = event.data;
+			const argsRaw = typeof data?.arguments === "string" ? data.arguments : JSON.stringify(data?.arguments ?? {});
+			toolLines.push(`[call] ${data?.name ?? "?"} ${argsRaw.slice(0, 500)}`);
+		} else if (event?.type === "tool/result") {
+			const data = event.data;
+			const output = typeof data?.output === "string" ? data.output : "";
+			const failure = data?.error ? " [ERROR]" : "";
+			toolLines.push(`[result]${failure} ${output.slice(0, 500)}`);
+		}
+	}
+	toolLines.reverse();
 	return [
 		`Review kind: ${kind}`,
 		`Signals: ${signal.toolCalls} tool calls, ${signal.userChars} user chars, ${signal.assistantChars} assistant chars.`,
+		`Recent tool activity (${toolLines.length}):`,
+		...toolLines,
 		"Return ONLY the structured JSON plan. Evidence is mandatory for every op.",
 		"",
 		...messages
 	].join("\n");
 }
 //#endregion
-export { Config, apply, inject, name };
+export { Config, apply, filterUnreadSkillOps, inject, name, shouldCompletionReview, sweepDeadSessionEntries };

package/lib/types/index.d.ts

@@ -11,7 +11,11 @@ export interface Config {
     reviewMode?: string;
     memoryInterval?: number;
     skillInterval?: number;
-    /** Tools the one-shot review subagent may use. Defaults include the Anchored Standard discovery pair. */
+    /**
+     * Tools the one-shot review subagent may use. Only actually-existing tools
+     * may be listed (the DSH tool catalog has `skill`; the `skill_search` /
+     * `skill_load` discovery pair does not exist on this platform).
+     */
     reviewToolAllow?: string[];
     reviewTimeoutMs?: number;
     executionTimeoutMs?: number;
@@ -20,7 +24,35 @@ export interface Config {
     reviewMaxDepth?: number;
     /** LLM provider for review subagents. Omit to inherit the deployment default route. */
     reviewProvider?: string;
+    /** Skill-review trigger: cadence (interval) | completion (once after a proven-long task) | both. */
+    skillReviewTrigger?: string;
+    /** Cumulative session tool calls before a session counts as proven-long for the completion channel. */
+    skillReviewCompletionMinToolCalls?: number;
 }
 export declare const Config: z<Config>;
 export declare function apply(ctx: Context, rawConfig: Config): void;
+/** Completion-channel decision: task finished normally AND the session is proven long. */
+export declare function shouldCompletionReview(reason: {
+    kind?: string;
+} | undefined, sessionToolCalls: number, minToolCalls: number): boolean;
+/**
+ * Remove every entry whose session is no longer live (rc.42 audit P1-10):
+ * `turnStarts` / `cumulativeToolCalls` / `completionInjected` are keyed by
+ * SessionId with no platform session-end hook to prune against, so they grew
+ * unbounded over a long-lived host. Works for maps and sets; returns the
+ * number of removed entries.
+ */
+export declare function sweepDeadSessionEntries<K>(entries: Map<K, unknown> | Set<K>, isAlive: (id: K) => boolean): number;
+/**
+ * Drop mutating ops whose target was not read this session, in place.
+ * Create is exempt (no read required to author a new skill). Covers the same
+ * mutating surface Hermes guards (edit/patch/write_file/remove_file), so a
+ * background review cannot blind-touch support files or edits of skills it
+ * never loaded. Returns the count of dropped ops so the plan event can report
+ * them as rejected.
+ */
+export declare function filterUnreadSkillOps(ops: Array<{
+    action?: string;
+    name?: string;
+}>, readNames: ReadonlySet<string>): number;
 //# sourceMappingURL=index.d.ts.map

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@lmzhen/dsh-evolution-review",
   "description": "Background review orchestration (community build)",
-  "version": "0.1.0-rc.9",
+  "version": "0.2.0-rc.1",
   "publishConfig": {
     "access": "public"
   },
@@ -33,27 +33,29 @@
   "license": "MIT",
   "dependencies": {
     "@deepseek-ai/schemastery": "^3.18.1",
-    "@lmzhen/dsh-evolution-core": "^0.1.0-rc.9",
-    "@lmzhen/dsh-evolution-plan-validator": "^0.1.0-rc.9"
+    "@lmzhen/dsh-evolution-core": "^0.2.0-rc.1",
+    "@lmzhen/dsh-evolution-plan-validator": "^0.2.0-rc.1"
   },
   "peerDependencies": {
     "@deepseek-ai/cordis": "^4.0.1",
-    "@deepseek-ai/dsh-agent": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-invariants": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-llm": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-session": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-tools": "^0.1.0-rc.6",
-    "@lmzhen/dsh-evolution-state": "^0.1.0-rc.9"
+    "@deepseek-ai/dsh-agent": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-invariants": "^0.1.1-rc.2",
+    "@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": "^0.2.0-rc.1"
   },
   "devDependencies": {
-    "@deepseek-ai/dsh-agent": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-agent-loop-testkit": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-invariants": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-llm": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-session": "^0.1.0-rc.6",
-    "@deepseek-ai/dsh-tools": "^0.1.0-rc.6",
-    "@lmzhen/dsh-evolution-core": "^0.1.0-rc.9",
-    "@lmzhen/dsh-evolution-plan-validator": "^0.1.0-rc.9",
-    "@lmzhen/dsh-evolution-state": "^0.1.0-rc.9"
+    "@deepseek-ai/dsh-agent": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-agent-loop-testkit": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-invariants": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-llm": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-session": "^0.1.1-rc.2",
+    "@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-core": "^0.2.0-rc.1",
+    "@lmzhen/dsh-evolution-plan-validator": "^0.2.0-rc.1",
+    "@lmzhen/dsh-evolution-state": "^0.2.0-rc.1"
   }
 }