dsh-mnemon-source-skills 0.5.9 → 0.5.10

package/README.md

@@ -61,3 +61,5 @@ The optional `./client` entry is an ordinary DSH Client plugin. This package own
 From a source checkout, install the declared dependencies and run `pnpm verify` to check the Source's behavior, the market's parsers and archive reader, and Source-backed page interactions.
 
 [Plugin development](https://github.com/omdsh-dev/dsh-mnemon/blob/main/docs/en/development/extensions.md) · [中文指南](https://github.com/omdsh-dev/dsh-mnemon/blob/main/docs/zh-CN/development/extensions.md)
+
+The optional Host `skills.autoAdopt` setting is off by default. It queues only new idle-review suggestions, requires complete paired text backtests without per-case or per-check regression, retries at most two rewrites, and uses Source-owned adoption under the storage lock. Source checks the exact draft/cases/results and current file again; cancellation is checked inside queued writes. Automatic activity and manual-review reasons are visible on each suggestion. See the [Skills guide](../../docs/en/guides/skills.md#optional-automatic-adoption) for limits and the distinction from real external-action validation.

package/lib/client.js

@@ -35,7 +35,7 @@ window.__ModuleLoader__.load({
 		function json(value) {
 			return JSON.parse(JSON.stringify(value ?? null));
 		}
-		const CONFLICT = /revision conflict|changed since it was read; reload/u;
+		const CONFLICT = /revision conflict|Skills Source changed since it was read; reload/u;
 		/**
 		* The Source's revision moves with every background review and run, so each
 		* write fetches it first and tries once more if a background write slipped in
@@ -127,6 +127,7 @@ window.__ModuleLoader__.load({
 					id,
 					content
 				}),
+				cancelRun: (id) => mutate("run-cancel", { id }),
 				adoptDraft: (id, root) => mutate("draft-adopt", root === void 0 ? { id } : {
 					id,
 					root
@@ -535,7 +536,34 @@ window.__ModuleLoader__.load({
 			"evolution.loadFailedBody": "请重试，或在左侧选择其他建议。",
 			"tests.model": "全局模型",
 			"tests.modelCard": "所选回测模型",
-			"tests.modelAuto": "跟随默认"
+			"tests.modelAuto": "跟随默认",
+			"draft.baseChanged": "原技能已变更，此草案已过期。请基于当前文件重新生成并回测。",
+			"draft.targetExists": "同名技能已存在。请重新生成为改进草案并回测，再确认采纳。",
+			"draft.refreshGuidance": "基于当前技能文件重新生成，保留最近的修改，并重新运行回测。",
+			"draft.regenerate": "重新生成并回测",
+			"draft.cancelBacktest": "取消回测",
+			"draft.interrupted": "回测已中断。可以重新回测，或在检查草案后手动采纳。",
+			"evolution.auto.title": "自动采纳",
+			"evolution.auto.history": "自动采纳活动记录",
+			"evolution.auto.attempt": "第 {count} 轮",
+			"evolution.auto.scores": "基线 {baseline} / 草案 {draft}",
+			"evolution.auto.stage.queued": "等待处理",
+			"evolution.auto.stage.drafting": "正在起草",
+			"evolution.auto.stage.backtesting": "正在对比回测",
+			"evolution.auto.stage.rewriting": "依据退步结果重写",
+			"evolution.auto.stage.adopted": "已自动采纳",
+			"evolution.auto.stage.paused": "已停止自动处理",
+			"evolution.auto.stage.review": "需要人工审查",
+			"evolution.auto.reason.disabled": "自动采纳或所需权限已关闭",
+			"evolution.auto.reason.no-cases": "没有启用的回测用例",
+			"evolution.auto.reason.case-limit": "启用的用例超过每轮上限；调整用例或上限后手动处理本建议",
+			"evolution.auto.reason.unmeasured": "回测未完整测出所有检查项，或存在图片依赖、错误、中断",
+			"evolution.auto.reason.regressed": "至少一个用例或原先通过的检查项出现退步",
+			"evolution.auto.reason.changed": "建议、草案或用例已变动",
+			"evolution.auto.reason.failed": "自动处理失败，保留人工处理入口",
+			"evolution.auto.reason.queue-full": "自动处理队列已满",
+			"evolution.auto.reason.model-unavailable": "无法确定起草或回测模型",
+			"evolution.auto.reason.interrupted": "Host 已停止，不自动重启此建议"
 		};
 		var en = {
 			"page.title": "Skills",
@@ -891,7 +919,34 @@ window.__ModuleLoader__.load({
 			"evolution.loadFailedBody": "Retry, or select another suggestion on the left.",
 			"tests.model": "Global model",
 			"tests.modelCard": "Selected model",
-			"tests.modelAuto": "Follow default"
+			"tests.modelAuto": "Follow default",
+			"draft.baseChanged": "The original skill changed. This draft is out of date; regenerate it against the current file and backtest it.",
+			"draft.targetExists": "A skill with this name now exists. Regenerate an update and backtest it before adopting.",
+			"draft.refreshGuidance": "Regenerate against the current skill file, preserving its latest changes, then run backtests.",
+			"draft.regenerate": "Regenerate and backtest",
+			"draft.cancelBacktest": "Cancel backtests",
+			"draft.interrupted": "Backtests were interrupted. Run them again, or review the draft before adopting manually.",
+			"evolution.auto.title": "Automatic adoption",
+			"evolution.auto.history": "Automatic adoption activity",
+			"evolution.auto.attempt": "Attempt {count}",
+			"evolution.auto.scores": "Baseline {baseline} / draft {draft}",
+			"evolution.auto.stage.queued": "Queued",
+			"evolution.auto.stage.drafting": "Drafting",
+			"evolution.auto.stage.backtesting": "Backtesting a pair",
+			"evolution.auto.stage.rewriting": "Rewriting after regression",
+			"evolution.auto.stage.adopted": "Automatically adopted",
+			"evolution.auto.stage.paused": "Automatic work stopped",
+			"evolution.auto.stage.review": "Manual review required",
+			"evolution.auto.reason.disabled": "Automatic adoption or its required permissions were disabled",
+			"evolution.auto.reason.no-cases": "No enabled evaluation cases",
+			"evolution.auto.reason.case-limit": "Enabled cases exceed the per-run limit; adjust cases or the limit and review this suggestion manually",
+			"evolution.auto.reason.unmeasured": "Some checks were not fully measured, or a case needs images, failed or was interrupted",
+			"evolution.auto.reason.regressed": "At least one case or previously passing check regressed",
+			"evolution.auto.reason.changed": "The suggestion, draft or evaluation cases changed",
+			"evolution.auto.reason.failed": "Automatic work failed; manual review remains available",
+			"evolution.auto.reason.queue-full": "Automatic queue is full",
+			"evolution.auto.reason.model-unavailable": "No resolved draft or backtest model",
+			"evolution.auto.reason.interrupted": "Host stopped; this suggestion will not restart automatically"
 		};
 		//#endregion
 		//#region \0source-css:/home/runner/work/dsh-mnemon/dsh-mnemon/plugins/dsh-mnemon-source-skills/presentation/page.module.css.mjs
@@ -3730,6 +3785,7 @@ window.__ModuleLoader__.load({
 					context.refresh();
 				} catch (error) {
 					context.fail(error);
+					context.refresh();
 				} finally {
 					setBusy(false);
 				}
@@ -3821,6 +3877,7 @@ window.__ModuleLoader__.load({
 						className: css.skWarn,
 						children: t("evolution.skillChanged")
 					}),
+					suggestion.autoAdoption !== void 0 && suggestion.autoAdoption.length > 0 && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(AutoAdoptionHistory, { events: suggestion.autoAdoption }),
 					suggestion.status === "drafting" && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Busy, { label: t("evolution.drafting") }),
 					suggestion.lastError !== void 0 && suggestion.status === "pending" && /* @__PURE__ */ (0, react_jsx_runtime.jsx)("p", {
 						className: css.inlineError,
@@ -3979,6 +4036,7 @@ window.__ModuleLoader__.load({
 					context.refresh();
 				} catch (error) {
 					context.fail(error);
+					context.refresh();
 				} finally {
 					setBusy(false);
 				}
@@ -4004,7 +4062,8 @@ window.__ModuleLoader__.load({
 				});
 				await context.client.linkRun(draft.id, started.runId);
 			});
-			const redraft = (note) => act(() => context.client.generateDraft({
+			const redraft = (note, backtest = false) => act(() => context.client.generateDraft({
+				...backtest ? { backtest: true } : {},
 				draftId: draft.id,
 				...draft.suggestionId === void 0 ? {} : { suggestionId: draft.suggestionId },
 				guidance: note,
@@ -4045,7 +4104,7 @@ window.__ModuleLoader__.load({
 							/* @__PURE__ */ (0, react_jsx_runtime.jsx)(_deepseek_ai_dsh_client_ui_primitives.Button, {
 								variant: "primary",
 								size: "sm",
-								disabled: !context.writable || busy || editing !== null || running || !originalAvailable,
+								disabled: !context.writable || busy || editing !== null || running || view.conflict !== void 0 || !originalAvailable,
 								onClick: () => regressed ? setConfirmAdopt(true) : void adopt(),
 								children: t("draft.adopt")
 							}),
@@ -4074,6 +4133,16 @@ window.__ModuleLoader__.load({
 						className: css.skWarn,
 						children: t(view.resolution === "origin-unavailable" ? "evolution.originUnavailable" : "evolution.skillGone")
 					}),
+					view.conflict !== void 0 && ready && /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+						className: css.skWarn,
+						role: "status",
+						children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("p", { children: t(view.conflict === "target-exists" ? "draft.targetExists" : "draft.baseChanged") }), originalAvailable && context.client.canAssist("skill-draft") && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(_deepseek_ai_dsh_client_ui_primitives.Button, {
+							size: "sm",
+							disabled: !context.writable || busy || running,
+							onClick: () => void redraft(t("draft.refreshGuidance"), true),
+							children: t("draft.regenerate")
+						})]
+					}),
 					draft.notes !== "" && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Clamp, { text: draft.notes }),
 					draft.changes.length > 0 && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(MoreList, {
 						items: draft.changes,
@@ -4082,37 +4151,53 @@ window.__ModuleLoader__.load({
 					/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 						className: css.skBacktest,
 						"data-tone": regressed ? "bad" : candidateScore !== void 0 && baselineScore !== void 0 && candidateScore > baselineScore ? "good" : void 0,
-						children: [running ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Busy, { label: t("draft.backtesting", {
-							done: progress(candidate) + progress(baseline),
-							total
-						}) }) : candidateScore !== void 0 && baselineScore !== void 0 ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("button", {
-							type: "button",
-							className: css.skBacktestLine,
-							"aria-expanded": details,
-							onClick: () => setDetails((value) => !value),
-							children: [
-								/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", { children: t("draft.backtest") }),
-								/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("strong", { children: [
-									baseline?.target.kind === "none" ? t("draft.withoutSkill") : t("draft.current"),
-									" ",
-									points(baselineScore),
-									" → ",
-									t("draft.candidate"),
-									" ",
-									points(candidateScore)
-								] }),
-								/* @__PURE__ */ (0, react_jsx_runtime.jsx)("em", { children: candidateScore === baselineScore ? t("draft.same") : `${candidateScore > baselineScore ? "+" : "−"}${Math.round(Math.abs(candidateScore - baselineScore) * 100)}` })
-							]
-						}) : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-							className: css.skMuted,
-							children: cases === 0 ? t("draft.noTests") : view.runs.length === 0 ? t("draft.notTested") : t("draft.notComparable")
-						}), !running && (cases ?? 0) > 0 && context.client.canAssist("skill-backtest") && ready && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(_deepseek_ai_dsh_client_ui_primitives.Button, {
-							variant: "ghost",
-							size: "sm",
-							disabled: !context.writable || busy || !originalAvailable,
-							onClick: () => void backtest(),
-							children: view.runs.length === 0 ? t("draft.runBacktest") : t("draft.rerun")
-						})]
+						children: [
+							running ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Busy, { label: t("draft.backtesting", {
+								done: progress(candidate) + progress(baseline),
+								total
+							}) }) : candidateScore !== void 0 && baselineScore !== void 0 ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("button", {
+								type: "button",
+								className: css.skBacktestLine,
+								"aria-expanded": details,
+								onClick: () => setDetails((value) => !value),
+								children: [
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", { children: t("draft.backtest") }),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("strong", { children: [
+										baseline?.target.kind === "none" ? t("draft.withoutSkill") : t("draft.current"),
+										" ",
+										points(baselineScore),
+										" → ",
+										t("draft.candidate"),
+										" ",
+										points(candidateScore)
+									] }),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("em", { children: candidateScore === baselineScore ? t("draft.same") : `${candidateScore > baselineScore ? "+" : "−"}${Math.round(Math.abs(candidateScore - baselineScore) * 100)}` })
+								]
+							}) : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+								className: css.skMuted,
+								children: cases === 0 ? t("draft.noTests") : view.runs.length === 0 ? t("draft.notTested") : t("draft.notComparable")
+							}),
+							running && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(_deepseek_ai_dsh_client_ui_primitives.Button, {
+								variant: "ghost",
+								size: "sm",
+								disabled: !context.writable || busy,
+								onClick: () => void act(async () => {
+									for (const run of view.runs.filter((item) => item.status === "running")) await context.client.cancelRun(run.id);
+								}),
+								children: t("draft.cancelBacktest")
+							}),
+							!running && (candidate?.status === "interrupted" || baseline?.status === "interrupted") && /* @__PURE__ */ (0, react_jsx_runtime.jsx)("p", {
+								role: "status",
+								children: t("draft.interrupted")
+							}),
+							!running && (cases ?? 0) > 0 && context.client.canAssist("skill-backtest") && ready && /* @__PURE__ */ (0, react_jsx_runtime.jsx)(_deepseek_ai_dsh_client_ui_primitives.Button, {
+								variant: "ghost",
+								size: "sm",
+								disabled: !context.writable || busy || !originalAvailable,
+								onClick: () => void backtest(),
+								children: view.runs.length === 0 ? t("draft.runBacktest") : t("draft.rerun")
+							})
+						]
 					}),
 					excluded > 0 && !running && /* @__PURE__ */ (0, react_jsx_runtime.jsx)("p", {
 						className: css.skMuted,
@@ -4208,6 +4293,35 @@ window.__ModuleLoader__.load({
 				]
 			});
 		}
+		function AutoAdoptionHistory(props) {
+			const t = useT();
+			const latest = props.events[props.events.length - 1];
+			const describe = (event) => /* @__PURE__ */ (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, { children: [
+				t(`evolution.auto.stage.${event.stage}`),
+				event.attempt > 0 ? ` · ${t("evolution.auto.attempt", { count: event.attempt })}` : "",
+				event.reason === void 0 ? "" : ` · ${t(`evolution.auto.reason.${event.reason}`)}`,
+				event.baselineScore === void 0 || event.draftScore === void 0 ? "" : ` · ${t("evolution.auto.scores", {
+					baseline: points(event.baselineScore),
+					draft: points(event.draftScore)
+				})}`,
+				event.detail === void 0 ? "" : ` · ${event.detail}`
+			] });
+			return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("section", {
+				"aria-label": t("evolution.auto.title"),
+				children: [/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("p", { children: [
+					/* @__PURE__ */ (0, react_jsx_runtime.jsx)("strong", { children: t("evolution.auto.title") }),
+					" · ",
+					describe(latest)
+				] }), /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Fold, {
+					title: t("evolution.auto.history"),
+					children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)("ol", { children: props.events.map((event, index) => /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("li", { children: [
+						stamp(event.at),
+						" · ",
+						describe(event)
+					] }, index)) })
+				})]
+			});
+		}
 		//#endregion
 		//#region src/contracts.ts
 		const SKILL_ROOT_KINDS = [

package/lib/contracts-CwhbQi-u.js

@@ -0,0 +1,119 @@
+//#region src/auto-adoption.ts
+/** Stable evaluation inputs, excluding presentation timestamps and provenance. Hash at the server boundary. */
+function skillCaseSignature(cases) {
+	return JSON.stringify([...cases].sort((a, b) => a.id < b.id ? -1 : a.id > b.id ? 1 : 0).map((entry) => [
+		entry.id,
+		entry.skillName,
+		entry.title,
+		entry.prompt,
+		entry.context ?? "",
+		entry.enabled,
+		entry.requiresVision === true,
+		entry.checks.map((check) => [
+			check.id,
+			check.text,
+			check.must
+		])
+	]));
+}
+/** Automatic decisions require full matching coverage, not the UI's scored intersection. */
+function compareAutomaticSkillRuns(cases, baseline, draft, casesHash) {
+	const unmeasured = { outcome: "unmeasured" };
+	if (cases.length === 0 || cases.some((entry) => !entry.enabled || entry.requiresVision === true || entry.checks.length === 0)) return unmeasured;
+	const ids = new Set(cases.map((entry) => entry.id));
+	if (ids.size !== cases.length || baseline === void 0 || draft === void 0 || baseline.model === void 0 || draft.model === void 0 || baseline.model.provider !== draft.model.provider || baseline.model.model !== draft.model.model || baseline.model.reasoningEffort !== draft.model.reasoningEffort) return unmeasured;
+	const complete = (run) => {
+		if (run.status !== "completed" || run.casesHash !== casesHash || run.errors || run.caseIds.length !== ids.size || new Set(run.caseIds).size !== ids.size || run.caseIds.some((id) => !ids.has(id)) || run.results.length !== ids.size || run.progress.done !== ids.size || run.progress.total !== ids.size) return;
+		const results = new Map(run.results.map((result) => [result.caseId, result]));
+		if (results.size !== ids.size) return;
+		for (const entry of cases) {
+			const result = results.get(entry.id);
+			if (result === void 0 || result.fullyGraded !== true || result.verdict === "error" || result.unmeasuredReason !== void 0 || result.error !== void 0 || !Number.isFinite(result.score)) return;
+			const checks = new Map(result.checks.map((check) => [check.checkId, check.pass]));
+			if (checks.size !== entry.checks.length || result.checks.length !== checks.size || entry.checks.some((check) => typeof checks.get(check.id) !== "boolean")) return;
+			const weight = entry.checks.reduce((sum, check) => sum + (check.must ? 2 : 1), 0);
+			const score = entry.checks.reduce((sum, check) => sum + (checks.get(check.id) ? check.must ? 2 : 1 : 0), 0) / weight;
+			const verdict = entry.checks.every((check) => !check.must || checks.get(check.id)) ? "pass" : "fail";
+			if (Math.abs(result.score - score) > 1e-8 || result.verdict !== verdict) return;
+		}
+		return results;
+	};
+	const before = complete(baseline);
+	const after = complete(draft);
+	if (before === void 0 || after === void 0) return unmeasured;
+	const baselineScore = cases.reduce((sum, entry) => sum + before.get(entry.id).score, 0) / cases.length;
+	const draftScore = cases.reduce((sum, entry) => sum + after.get(entry.id).score, 0) / cases.length;
+	return {
+		outcome: cases.some((entry) => {
+			const left = before.get(entry.id);
+			const right = after.get(entry.id);
+			const checks = new Map(right.checks.map((check) => [check.checkId, check.pass]));
+			return right.score + 1e-8 < left.score || left.checks.some((check) => check.pass && !checks.get(check.checkId));
+		}) ? "regressed" : "not-worse",
+		baselineScore,
+		draftScore
+	};
+}
+//#endregion
+//#region src/contracts.ts
+const SKILL_ROOT_KINDS = [
+	"project-dsh",
+	"project-agents",
+	"user-dsh",
+	"user-agents"
+];
+const SKILL_ROOT_RANK = {
+	"project-dsh": 100,
+	"project-agents": 200,
+	"user-dsh": 400,
+	"user-agents": 500
+};
+const EVOLUTION_SIGNALS = [
+	"correction",
+	"directive",
+	"tool-failure",
+	"approach-change",
+	"procedure",
+	"skill-gap",
+	"skill-misleading",
+	"success"
+];
+const SUGGESTION_STATUSES = [
+	"pending",
+	"drafting",
+	"review",
+	"adopted",
+	"dismissed",
+	"expired",
+	"invalidated"
+];
+const DEFAULT_SKILLS_POLICY = Object.freeze({
+	allowCreate: true,
+	allowUpdate: true,
+	minConfidence: .6,
+	expiryDays: 7,
+	maxPendingPerSession: 5
+});
+const AUTO_ADOPTION_STAGES = [
+	"queued",
+	"drafting",
+	"backtesting",
+	"rewriting",
+	"adopted",
+	"paused",
+	"review"
+];
+const AUTO_ADOPTION_REASONS = [
+	"disabled",
+	"no-cases",
+	"case-limit",
+	"unmeasured",
+	"regressed",
+	"changed",
+	"failed",
+	"queue-full",
+	"model-unavailable",
+	"interrupted"
+];
+//#endregion
+export { SKILL_ROOT_KINDS as a, compareAutomaticSkillRuns as c, EVOLUTION_SIGNALS as i, skillCaseSignature as l, AUTO_ADOPTION_STAGES as n, SKILL_ROOT_RANK as o, DEFAULT_SKILLS_POLICY as r, SUGGESTION_STATUSES as s, AUTO_ADOPTION_REASONS as t };

package/lib/contracts.d.ts

@@ -20,6 +20,16 @@ interface DetectedSource {
   label: string;
 }
 //#endregion
+//#region src/auto-adoption.d.ts
+/** Stable evaluation inputs, excluding presentation timestamps and provenance. Hash at the server boundary. */
+declare function skillCaseSignature(cases: readonly EvalCase[]): string;
+/** Automatic decisions require full matching coverage, not the UI's scored intersection. */
+declare function compareAutomaticSkillRuns(cases: readonly EvalCase[], baseline: BacktestRun | undefined, draft: BacktestRun | undefined, casesHash: string): {
+  outcome: 'unmeasured' | 'regressed' | 'not-worse';
+  baselineScore?: number;
+  draftScore?: number;
+};
+//#endregion
 //#region src/contracts.d.ts
 /** The DSH skill roots this Source reads and writes, in DSH's precedence order. */
 type SkillRootKind = 'project-dsh' | 'project-agents' | 'user-dsh' | 'user-agents';
@@ -192,6 +202,7 @@ interface SkillSuggestion {
   status: SuggestionStatus;
   draftId?: string;
   lastError?: string;
+  autoAdoption?: AutoAdoptionEvent[];
   createdAt: string;
   updatedAt: string;
   expiresAt: string;
@@ -292,6 +303,8 @@ interface CaseResult {
   error?: string;
   /** Kept under verdict=error so older readers also exclude it from scores. */
   unmeasuredReason?: 'requires-vision';
+  /** Every expected judge check was explicitly returned; absent on older runs. */
+  fullyGraded?: boolean;
 }
 type RunStatus = 'running' | 'completed' | 'failed' | 'interrupted';
 interface BacktestRun {
@@ -304,6 +317,11 @@ interface BacktestRun {
     total: number;
   };
   caseIds: string[];
+  /** Source-verified case inputs at start; absent on older runs. */
+  casesHash?: string;
+  model?: ModelRoute & {
+    reasoningEffort?: string;
+  };
   results: CaseResult[];
   /** Over the graded cases; absent when none could be graded. */
   passRate?: number;
@@ -330,6 +348,10 @@ interface ReviewLogEntry {
   error?: string;
   /** The conversation turn the reviewed window ended with. */
   turn?: number;
+  autoAdoptions?: Array<AutoAdoptionEvent & {
+    suggestionId: string;
+    skillName: string;
+  }>;
   /** The suggestions this review created or strengthened. */
   suggestionIds?: string[];
 }
@@ -433,6 +455,8 @@ interface ReviewRecordResult {
   reviewId: string;
   trajectoryId?: string;
   suggestionIds: string[];
+  /** Newly inserted ids only; older Sources omit this capability. */
+  createdSuggestionIds?: string[];
   inserted: number;
   updated: number;
   casesAdded: number;
@@ -487,6 +511,8 @@ interface SuggestionView {
 }
 interface DraftView {
   draft: SkillDraft;
+  /** File changes need a new draft, never a retry with a newer revision. */
+  conflict?: 'base-changed' | 'target-exists';
   /** The file the draft replaces, as it is now. */
   base?: SkillDocument;
   resolution?: SkillResolution;
@@ -637,5 +663,27 @@ interface MarketInstall {
   files: number;
   installedAt: string;
 }
+declare const AUTO_ADOPTION_STAGES: readonly ["queued", "drafting", "backtesting", "rewriting", "adopted", "paused", "review"];
+declare const AUTO_ADOPTION_REASONS: readonly ["disabled", "no-cases", "case-limit", "unmeasured", "regressed", "changed", "failed", "queue-full", "model-unavailable", "interrupted"];
+interface AutoAdoptionEvent {
+  at: string;
+  stage: typeof AUTO_ADOPTION_STAGES[number];
+  attempt: number;
+  reason?: typeof AUTO_ADOPTION_REASONS[number];
+  detail?: string;
+  baselineScore?: number;
+  draftScore?: number;
+}
+type DraftExpectation = {
+  id: string;
+  hash: string;
+} | null;
+interface AutomaticAdoption {
+  attempt: number;
+  draftHash: string;
+  baselineRunId: string;
+  draftRunId: string;
+  casesHash: string;
+}
 //#endregion
-export { BacktestRun, CaseResult, CatalogView, DEFAULT_SKILLS_POLICY, type DetectedSource, DraftPreferences, DraftPreferencesView, DraftStatus, DraftView, EVOLUTION_SIGNALS, EvalCase, EvalCheck, EvidencePack, EvidenceReference, EvolutionSignal, LocalizedText, MarketAudit, MarketEntry, MarketInstall, MarketInstallState, MarketItem, MarketPreview, MarketRegistry, MarketRegistryFile, MarketSearchResult, MarketSource, type MarketSourceKind, MarketView, ModelRoute, ReviewLogEntry, ReviewRecordInput, ReviewRecordResult, ReviewScore, RunStatus, RunTarget, RunTargetKind, SKILL_ROOT_KINDS, SKILL_ROOT_RANK, SUGGESTION_STATUSES, SkillCatalog, SkillDocument, SkillDraft, SkillInvocation, SkillOrigin, SkillProblem, SkillResolution, SkillRoot, SkillRootKind, SkillSuggestion, SkillSummary, SkillVersion, SkillVersionReason, SkillsOverview, SkillsPolicy, SuggestionStatus, SuggestionView, Trajectory, TrajectoryAnnotation, TrajectoryStats, TrajectoryStep, TrajectoryStepKind, TrajectorySummary, TrashEntry, TrashRestoreInput, TrashRestorePreview };
+export { AUTO_ADOPTION_REASONS, AUTO_ADOPTION_STAGES, AutoAdoptionEvent, AutomaticAdoption, BacktestRun, CaseResult, CatalogView, DEFAULT_SKILLS_POLICY, type DetectedSource, DraftExpectation, DraftPreferences, DraftPreferencesView, DraftStatus, DraftView, EVOLUTION_SIGNALS, EvalCase, EvalCheck, EvidencePack, EvidenceReference, EvolutionSignal, LocalizedText, MarketAudit, MarketEntry, MarketInstall, MarketInstallState, MarketItem, MarketPreview, MarketRegistry, MarketRegistryFile, MarketSearchResult, MarketSource, type MarketSourceKind, MarketView, ModelRoute, ReviewLogEntry, ReviewRecordInput, ReviewRecordResult, ReviewScore, RunStatus, RunTarget, RunTargetKind, SKILL_ROOT_KINDS, SKILL_ROOT_RANK, SUGGESTION_STATUSES, SkillCatalog, SkillDocument, SkillDraft, SkillInvocation, SkillOrigin, SkillProblem, SkillResolution, SkillRoot, SkillRootKind, SkillSuggestion, SkillSummary, SkillVersion, SkillVersionReason, SkillsOverview, SkillsPolicy, SuggestionStatus, SuggestionView, Trajectory, TrajectoryAnnotation, TrajectoryStats, TrajectoryStep, TrajectoryStepKind, TrajectorySummary, TrashEntry, TrashRestoreInput, TrashRestorePreview, compareAutomaticSkillRuns, skillCaseSignature };

package/lib/contracts.js

@@ -1,41 +1,2 @@
-//#region src/contracts.ts
-const SKILL_ROOT_KINDS = [
-	"project-dsh",
-	"project-agents",
-	"user-dsh",
-	"user-agents"
-];
-const SKILL_ROOT_RANK = {
-	"project-dsh": 100,
-	"project-agents": 200,
-	"user-dsh": 400,
-	"user-agents": 500
-};
-const EVOLUTION_SIGNALS = [
-	"correction",
-	"directive",
-	"tool-failure",
-	"approach-change",
-	"procedure",
-	"skill-gap",
-	"skill-misleading",
-	"success"
-];
-const SUGGESTION_STATUSES = [
-	"pending",
-	"drafting",
-	"review",
-	"adopted",
-	"dismissed",
-	"expired",
-	"invalidated"
-];
-const DEFAULT_SKILLS_POLICY = Object.freeze({
-	allowCreate: true,
-	allowUpdate: true,
-	minConfidence: .6,
-	expiryDays: 7,
-	maxPendingPerSession: 5
-});
-//#endregion
-export { DEFAULT_SKILLS_POLICY, EVOLUTION_SIGNALS, SKILL_ROOT_KINDS, SKILL_ROOT_RANK, SUGGESTION_STATUSES };
+import { a as SKILL_ROOT_KINDS, c as compareAutomaticSkillRuns, i as EVOLUTION_SIGNALS, l as skillCaseSignature, n as AUTO_ADOPTION_STAGES, o as SKILL_ROOT_RANK, r as DEFAULT_SKILLS_POLICY, s as SUGGESTION_STATUSES, t as AUTO_ADOPTION_REASONS } from "./contracts-CwhbQi-u.js";
+export { AUTO_ADOPTION_REASONS, AUTO_ADOPTION_STAGES, DEFAULT_SKILLS_POLICY, EVOLUTION_SIGNALS, SKILL_ROOT_KINDS, SKILL_ROOT_RANK, SUGGESTION_STATUSES, compareAutomaticSkillRuns, skillCaseSignature };

package/lib/index.js

@@ -1,4 +1,4 @@
-import { DEFAULT_SKILLS_POLICY, EVOLUTION_SIGNALS, SKILL_ROOT_KINDS, SKILL_ROOT_RANK } from "./contracts.js";
+import { a as SKILL_ROOT_KINDS, c as compareAutomaticSkillRuns, i as EVOLUTION_SIGNALS, l as skillCaseSignature, n as AUTO_ADOPTION_STAGES, o as SKILL_ROOT_RANK, r as DEFAULT_SKILLS_POLICY, t as AUTO_ADOPTION_REASONS } from "./contracts-CwhbQi-u.js";
 import { defineMemoryPlugin, defineMemorySource, installMemory, memoryConfigurationDigest, memoryInputRecord, memoryInputStringArray, memoryInputText, withMemoryStorageLock } from "dsh-mnemon/extension-sdk";
 import { homedir } from "node:os";
 import { basename, dirname, isAbsolute, join, relative, resolve, sep } from "node:path";
@@ -837,15 +837,39 @@ function orphanedDraft(suggestion) {
 */
 function expireSuggestions(ledger, now) {
 	let changed = false;
-	for (const suggestion of ledger.suggestions) if (suggestion.status === "pending" && Date.parse(suggestion.expiresAt) < now.getTime()) {
-		suggestion.status = "expired";
-		suggestion.updatedAt = now.toISOString();
-		changed = true;
-	} else if (suggestion.status === "drafting" && orphanedDraft(suggestion)) {
-		suggestion.status = "pending";
-		suggestion.updatedAt = now.toISOString();
-		suggestion.lastError = ORPHAN_DRAFT_ERROR;
-		changed = true;
+	for (const suggestion of ledger.suggestions) {
+		const lastAuto = suggestion.autoAdoption?.at(-1);
+		if (lastAuto !== void 0 && [
+			"queued",
+			"drafting",
+			"backtesting",
+			"rewriting"
+		].includes(lastAuto.stage) && (!Number.isFinite(Date.parse(lastAuto.at)) || Date.parse(lastAuto.at) < PROCESS_STARTED_AT)) {
+			const event = {
+				at: now.toISOString(),
+				stage: "paused",
+				attempt: lastAuto.attempt,
+				reason: "interrupted"
+			};
+			suggestion.autoAdoption = [...suggestion.autoAdoption, event].slice(-16);
+			const review = ledger.reviews.find((entry) => entry.id === suggestion.reviewId);
+			if (review !== void 0) review.autoAdoptions = [...review.autoAdoptions ?? [], {
+				...event,
+				suggestionId: suggestion.id,
+				skillName: suggestion.skillName
+			}].slice(-160);
+			changed = true;
+		}
+		if (suggestion.status === "pending" && Date.parse(suggestion.expiresAt) < now.getTime()) {
+			suggestion.status = "expired";
+			suggestion.updatedAt = now.toISOString();
+			changed = true;
+		} else if (suggestion.status === "drafting" && orphanedDraft(suggestion)) {
+			suggestion.status = "pending";
+			suggestion.updatedAt = now.toISOString();
+			suggestion.lastError = ORPHAN_DRAFT_ERROR;
+			changed = true;
+		}
 	}
 	return changed;
 }
@@ -900,6 +924,7 @@ function admitReview(context, input) {
 	const steps = (values) => [...new Set(values.filter((value) => Number.isInteger(value) && stepIndexes.has(value)))].sort((a, b) => a - b);
 	const skipped = [];
 	const allowed = (action) => (action === "create" ? policy.allowCreate : policy.allowUpdate) !== false;
+	const createdSuggestionIds = [];
 	let inserted = 0;
 	let updated = 0;
 	let casesAdded = 0;
@@ -1038,6 +1063,7 @@ function admitReview(context, input) {
 			expiresAt: new Date(now.getTime() + policy.expiryDays * 24 * 60 * 60 * 1e3).toISOString()
 		};
 		ledger.suggestions.unshift(suggestion);
+		createdSuggestionIds.push(suggestion.id);
 		touched.push(suggestion);
 		inserted++;
 	}
@@ -1151,6 +1177,7 @@ function admitReview(context, input) {
 			reviewId,
 			trajectoryId,
 			suggestionIds: touched.map((suggestion) => suggestion.id),
+			createdSuggestionIds,
 			inserted,
 			updated,
 			casesAdded,
@@ -3266,15 +3293,17 @@ var SkillsService = class {
 		const ledger = this.store.ledger();
 		return ledger.drafts.filter((draft) => (filter.suggestionId === void 0 || draft.suggestionId === filter.suggestionId) && (filter.skill === void 0 || draft.skillName === filter.skill) && (filter.path === void 0 || this.draftOriginal(ledger, draft).skill?.path === filter.path));
 	}
-	draft(_workspace, id) {
+	draft(workspace, id) {
 		const ledger = this.store.ledger();
 		const draft = ledger.drafts.find((candidate) => candidate.id === id);
 		if (draft === void 0) throw new Error("draft not found: " + id);
 		const original = draft.action === "update" ? this.draftOriginal(ledger, draft) : void 0;
 		const base = original?.skill;
+		const conflict = draft.action === "update" ? base !== void 0 && base.hash !== draft.baseHash ? "base-changed" : void 0 : activeSkill(this.catalog(workspace), draft.skillName) !== void 0 ? "target-exists" : void 0;
 		const now = this.clock().getTime();
 		return {
 			draft,
+			...conflict === void 0 ? {} : { conflict },
 			...original === void 0 ? {} : { resolution: original.resolution },
 			...base === void 0 ? {} : { base: readSkillDocument(base) },
 			runs: draft.runIds.flatMap((id) => {
@@ -3648,7 +3677,9 @@ var SkillsService = class {
 	}
 	/** The Host marks a suggestion `drafting` before a task Agent writes, and back to `pending` if it fails. */
 	suggestionStatus(input) {
-		return this.withSuggestion(input.id, (suggestion) => {
+		return this.withSuggestion(input.id, (suggestion, ledger) => {
+			input.signal?.throwIfAborted();
+			this.requireExpectedDraft(ledger, suggestion, input.expectedDraft);
 			if (input.status === "drafting") {
 				if (suggestion.status !== "pending" && suggestion.status !== "review") throw new Error("this suggestion is " + suggestion.status + "; only an open suggestion can be drafted");
 				suggestion.status = "drafting";
@@ -3661,6 +3692,28 @@ var SkillsService = class {
 			return suggestion;
 		});
 	}
+	requireExpectedDraft(ledger, suggestion, expected) {
+		if (expected === void 0) return;
+		const current = ledger.drafts.find((draft) => draft.id === suggestion?.draftId);
+		if (!(expected === null ? suggestion !== void 0 && suggestion.draftId === void 0 : current?.id === expected.id && current.status === "ready" && sha256(current.content) === expected.hash)) throw new Error("The suggestion or draft changed while this draft was being generated; leave it for manual review");
+	}
+	recordAutoAdoption(id, input) {
+		return this.withSuggestion(id, (suggestion, ledger) => {
+			const event = {
+				...input,
+				...input.detail === void 0 ? {} : { detail: clip(input.detail, 500) },
+				at: this.clock().toISOString()
+			};
+			suggestion.autoAdoption = [...suggestion.autoAdoption ?? [], event].slice(-16);
+			const review = ledger.reviews.find((entry) => entry.id === suggestion.reviewId);
+			if (review !== void 0) review.autoAdoptions = [...review.autoAdoptions ?? [], {
+				...event,
+				suggestionId: id,
+				skillName: suggestion.skillName
+			}].slice(-160);
+			return suggestion;
+		});
+	}
 	pinTrajectory(id, pinned) {
 		return this.write(() => {
 			const ledger = this.store.ledger();
@@ -3783,6 +3836,7 @@ var SkillsService = class {
 	}
 	recordDraft(workspace, input) {
 		return this.write(() => {
+			input.signal?.throwIfAborted();
 			validateSkillContent(input.content, input.skillName);
 			const ledger = this.store.ledger();
 			const catalog = this.catalog(workspace);
@@ -3793,8 +3847,10 @@ var SkillsService = class {
 			const suggestion = input.suggestionId === void 0 ? void 0 : ledger.suggestions.find((candidate) => candidate.id === input.suggestionId);
 			if (input.suggestionId !== void 0 && suggestion === void 0) throw new Error("suggestion not found: " + input.suggestionId);
 			if (suggestion !== void 0 && suggestion.skillName !== input.skillName) throw new Error("suggestion is for another skill");
+			this.requireExpectedDraft(ledger, suggestion, input.expectedDraft);
 			const original = prior?.action === "update" ? this.draftOriginal(ledger, prior) : suggestion?.action === "update" ? this.suggestionOriginal(ledger, suggestion) : void 0;
 			const loaded = original === void 0 ? activeSkill(catalog, input.skillName) : this.requireOriginal(original);
+			if (input.expectedBaseHash !== void 0 && (loaded?.hash ?? null) !== input.expectedBaseHash) throw new Error("The skill changed while this draft was being generated; regenerate it from the current file");
 			const origin = loaded === void 0 ? void 0 : original?.origin ?? skillOrigin(catalog, loaded);
 			for (const draft of ledger.drafts) if (draft.status === "ready" && draft.skillName === input.skillName && draft.suggestionId === input.suggestionId && (loaded === void 0 ? draft.action === "create" : this.draftOriginal(ledger, draft).skill?.path === loaded.path)) {
 				draft.status = "discarded";
@@ -3848,6 +3904,19 @@ var SkillsService = class {
 			const ledger = this.store.ledger();
 			const draft = ledger.drafts.find((candidate) => candidate.id === id);
 			if (draft === void 0 || draft.status !== "ready") throw new Error("only a draft waiting for review can be adopted");
+			let automaticResult;
+			if (input.automatic !== void 0) {
+				const proof = input.automatic;
+				const suggestion = ledger.suggestions.find((entry) => entry.id === draft.suggestionId);
+				const evals = this.store.evals();
+				const cases = evals.cases.filter((entry) => entry.skillName === draft.skillName && entry.enabled);
+				const casesHash = sha256(skillCaseSignature(cases));
+				const before = evals.runs.find((run) => run.id === proof.baselineRunId);
+				const after = evals.runs.find((run) => run.id === proof.draftRunId);
+				automaticResult = compareAutomaticSkillRuns(cases, before, after, casesHash);
+				if (suggestion?.status !== "review" || suggestion.draftId !== draft.id || Date.parse(suggestion.expiresAt) <= this.clock().getTime() || proof.draftHash !== sha256(draft.content) || casesHash !== proof.casesHash || before?.skillName !== draft.skillName || after?.skillName !== draft.skillName || !(Date.parse(before.startedAt) >= Date.parse(draft.createdAt)) || !(Date.parse(after.startedAt) >= Date.parse(draft.createdAt)) || after.target.kind !== "draft" || after.target.draftId !== draft.id || after.target.hash !== proof.draftHash || (draft.action === "update" ? before.target.kind !== "current" || before.target.hash !== draft.baseHash : before.target.kind !== "none") || automaticResult.outcome !== "not-worse") throw new Error("The draft, cases, or complete matching backtest results changed; automatic adoption requires manual review");
+			}
+			input.signal?.throwIfAborted();
 			validateSkillContent(draft.content, draft.skillName);
 			const catalog = this.catalog(workspace);
 			let key;
@@ -3885,6 +3954,22 @@ var SkillsService = class {
 			if (suggestion !== void 0) {
 				suggestion.status = "adopted";
 				suggestion.updatedAt = now;
+				if (input.automatic !== void 0 && automaticResult !== void 0) {
+					const event = {
+						at: now,
+						stage: "adopted",
+						attempt: input.automatic.attempt,
+						...automaticResult.baselineScore === void 0 ? {} : { baselineScore: automaticResult.baselineScore },
+						...automaticResult.draftScore === void 0 ? {} : { draftScore: automaticResult.draftScore }
+					};
+					suggestion.autoAdoption = [...suggestion.autoAdoption ?? [], event].slice(-16);
+					const review = ledger.reviews.find((entry) => entry.id === suggestion.reviewId);
+					if (review !== void 0) review.autoAdoptions = [...review.autoAdoptions ?? [], {
+						...event,
+						suggestionId: suggestion.id,
+						skillName: suggestion.skillName
+					}].slice(-160);
+				}
 			}
 			this.store.saveLedger(ledger);
 			return {
@@ -3945,6 +4030,26 @@ var SkillsService = class {
 			return { reclaimed: Math.max(0, before - after) };
 		});
 	}
+	/** Persist terminal state so expired tasks cannot later revive themselves. */
+	reclaimStaleRuns() {
+		return this.write(() => {
+			const evals = this.store.evals();
+			const now = this.clock();
+			let reclaimed = 0;
+			for (const run of evals.runs) {
+				if (run.status !== "running" || Number.isFinite(Date.parse(run.startedAt)) && now.getTime() - Date.parse(run.startedAt) <= RUN_STALE_MS) continue;
+				run.status = "interrupted";
+				run.finishedAt = now.toISOString();
+				run.error = "Backtest exceeded its 45 minute deadline; run it again.";
+				reclaimed++;
+			}
+			if (reclaimed > 0) {
+				this.store.saveEvals(evals);
+				this.touch();
+			}
+			return { reclaimed };
+		});
+	}
 	/** The content a run tests, resolved from its target. */
 	targetContent(workspace, skillName, target) {
 		if (target.kind === "none") return { target: { kind: "none" } };
@@ -3990,9 +4095,12 @@ var SkillsService = class {
 	}
 	startRun(input) {
 		return this.write(() => {
+			input.signal?.throwIfAborted();
 			const evals = this.store.evals();
 			const cases = input.caseIds.map((id) => evals.cases.find((entry) => entry.id === id && entry.skillName === input.skillName));
 			if (cases.length === 0 || cases.some((entry) => entry === void 0)) throw new Error("a run needs at least one case of this skill");
+			const casesHash = sha256(skillCaseSignature(cases));
+			if (input.expectedCasesHash !== void 0 && input.expectedCasesHash !== casesHash) throw new Error("Eval cases changed before this backtest started");
 			const run = {
 				id: randomUUID(),
 				skillName: input.skillName,
@@ -4003,6 +4111,8 @@ var SkillsService = class {
 					total: cases.length
 				},
 				caseIds: input.caseIds,
+				casesHash,
+				...input.model === void 0 ? {} : { model: input.model },
 				results: [],
 				startedAt: this.clock().toISOString()
 			};
@@ -4029,6 +4139,7 @@ var SkillsService = class {
 	}
 	progressRun(input) {
 		return this.write(() => {
+			input.signal?.throwIfAborted();
 			const evals = this.store.evals();
 			const run = evals.runs.find((candidate) => candidate.id === input.id);
 			if (run === void 0 || run.status !== "running") throw new Error("run is not running: " + input.id);
@@ -4055,9 +4166,11 @@ var SkillsService = class {
 	}
 	finishRun(input) {
 		return this.write(() => {
+			input.signal?.throwIfAborted();
 			const evals = this.store.evals();
 			const run = evals.runs.find((candidate) => candidate.id === input.id);
 			if (run === void 0) throw new Error("run not found: " + input.id);
+			if (run.status !== "running") return run;
 			run.status = input.status;
 			run.finishedAt = this.clock().toISOString();
 			if (input.error !== void 0) run.error = clip(input.error, 500);
@@ -4551,6 +4664,55 @@ function draftModelInput(value) {
 		model: memoryInputText(picked.model, "model.model", 200)
 	};
 }
+function backtestModel(value) {
+	if (value === void 0) return {};
+	const model = draftModelInput(value);
+	if (model === null) throw new Error("a backtest model must name its provider and model");
+	const input = memoryInputRecord(value, "model");
+	return { model: {
+		...model,
+		...typeof input.reasoningEffort === "string" ? { reasoningEffort: memoryInputText(input.reasoningEffort, "reasoningEffort", 100) } : {}
+	} };
+}
+function draftExpectation(value) {
+	if (value === void 0) return {};
+	if (value === null) return { expectedDraft: null };
+	const expected = memoryInputRecord(value, "expectedDraft");
+	return { expectedDraft: {
+		id: memoryInputText(expected.id, "expectedDraft.id", 100),
+		hash: memoryInputText(expected.hash, "expectedDraft.hash", 100)
+	} };
+}
+function automaticAdoption(value) {
+	if (value === void 0) return {};
+	const input = memoryInputRecord(value, "automatic");
+	if (!Number.isInteger(input.attempt) || Number(input.attempt) < 1 || Number(input.attempt) > 3) throw new Error("automatic adoption attempt must be 1..3");
+	return { automatic: {
+		attempt: Number(input.attempt),
+		draftHash: memoryInputText(input.draftHash, "draftHash", 100),
+		baselineRunId: memoryInputText(input.baselineRunId, "baselineRunId", 100),
+		draftRunId: memoryInputText(input.draftRunId, "draftRunId", 100),
+		casesHash: memoryInputText(input.casesHash, "casesHash", 100)
+	} };
+}
+function autoAdoptionEvent(value) {
+	if (!AUTO_ADOPTION_STAGES.includes(value.stage)) throw new Error("unknown automatic adoption stage");
+	if (value.reason !== void 0 && !AUTO_ADOPTION_REASONS.includes(value.reason)) throw new Error("unknown automatic adoption reason");
+	if (!Number.isInteger(value.attempt) || Number(value.attempt) < 0 || Number(value.attempt) > 3) throw new Error("automatic adoption attempt must be 0..3");
+	const score = (key) => {
+		if (value[key] === void 0) return {};
+		if (typeof value[key] !== "number" || !Number.isFinite(value[key]) || value[key] < 0 || value[key] > 1) throw new Error("invalid automatic adoption score");
+		return { [key]: value[key] };
+	};
+	return {
+		stage: value.stage,
+		attempt: Number(value.attempt),
+		...value.reason === void 0 ? {} : { reason: value.reason },
+		...typeof value.detail === "string" ? { detail: value.detail } : {},
+		...score("baselineScore"),
+		...score("draftScore")
+	};
+}
 function caseInput(value, skillName) {
 	const input = memoryInputRecord(value, "case");
 	const checks = input.checks;
@@ -4610,7 +4772,8 @@ function caseResult(value) {
 		answer: typeof input.answer === "string" ? input.answer : "",
 		...typeof input.summary === "string" ? { summary: input.summary } : {},
 		...typeof input.error === "string" ? { error: input.error } : {},
-		...unmeasuredReason === void 0 ? {} : { unmeasuredReason }
+		...unmeasuredReason === void 0 ? {} : { unmeasuredReason },
+		...typeof input.fullyGraded === "boolean" ? { fullyGraded: input.fullyGraded } : {}
 	};
 }
 function signal(value) {
@@ -4777,6 +4940,11 @@ function createSkillsMemorySource(config = {}) {
 				...effective.agentsHome === void 0 ? {} : { agentsHome: effective.agentsHome }
 			});
 			service.reclaimOrphanDrafts().catch(() => void 0);
+			const recovery = service.reclaimStaleRuns();
+			const reaper = setInterval(() => {
+				service.reclaimStaleRuns().catch(() => void 0);
+			}, 6e4);
+			reaper.unref?.();
 			const facts = (scope, management) => management ? {
 				sourceInstanceKey: context.sourceInstanceKey,
 				sourceTypeId: SKILLS_SOURCE_TYPE_ID,
@@ -4798,6 +4966,9 @@ function createSkillsMemorySource(config = {}) {
 				hints: { reason: "management-only" }
 			};
 			return {
+				dispose() {
+					clearInterval(reaper);
+				},
 				facts(request) {
 					return facts(request.scope, request.scenario.startsWith("management."));
 				},
@@ -4805,6 +4976,7 @@ function createSkillsMemorySource(config = {}) {
 					return { fragments: [] };
 				},
 				async manage(request) {
+					await recovery;
 					const root = workspace(request.scope);
 					const input = request.input ?? {};
 					const fields = typeof input === "object" && input !== null && !Array.isArray(input) ? input : {};
@@ -4931,12 +5103,17 @@ function createSkillsMemorySource(config = {}) {
 							const status = fields.status;
 							if (status !== "drafting" && status !== "pending") throw new Error("status must be drafting or pending");
 							value = await service.suggestionStatus({
+								...request.signal === void 0 ? {} : { signal: request.signal },
 								id: memoryInputText(fields.id, "id", 100),
 								status,
+								...draftExpectation(fields.expectedDraft),
 								...typeof fields.error === "string" ? { error: fields.error } : {}
 							});
 							break;
 						}
+						case "auto-adoption-record":
+							value = await service.recordAutoAdoption(memoryInputText(fields.id, "id", 100), autoAdoptionEvent(fields));
+							break;
 						case "trajectory-pin":
 							value = await service.pinTrajectory(memoryInputText(fields.id, "id", 100), bool(fields.pinned, "pinned"));
 							break;
@@ -4962,6 +5139,9 @@ function createSkillsMemorySource(config = {}) {
 						}
 						case "draft-record":
 							value = await service.recordDraft(root, {
+								...request.signal === void 0 ? {} : { signal: request.signal },
+								...draftExpectation(fields.expectedDraft),
+								...fields.expectedBaseHash === void 0 ? {} : { expectedBaseHash: fields.expectedBaseHash === null ? null : memoryInputText(fields.expectedBaseHash, "expectedBaseHash", 100) },
 								...optionalText(fields.suggestionId, "suggestionId", 100) === void 0 ? {} : { suggestionId: optionalText(fields.suggestionId, "suggestionId", 100) },
 								...optionalText(fields.draftId, "draftId", 100) === void 0 ? {} : { draftId: optionalText(fields.draftId, "draftId", 100) },
 								skillName: memoryInputText(fields.skillName, "skillName", 64),
@@ -4979,7 +5159,11 @@ function createSkillsMemorySource(config = {}) {
 							});
 							break;
 						case "draft-adopt":
-							value = await service.adoptDraft(root, memoryInputText(fields.id, "id", 100), { ...rootKind(fields.root) === void 0 ? {} : { root: rootKind(fields.root) } });
+							value = await service.adoptDraft(root, memoryInputText(fields.id, "id", 100), {
+								...request.signal === void 0 ? {} : { signal: request.signal },
+								...automaticAdoption(fields.automatic),
+								...rootKind(fields.root) === void 0 ? {} : { root: rootKind(fields.root) }
+							});
 							break;
 						case "draft-discard":
 							value = await service.discardDraft(memoryInputText(fields.id, "id", 100));
@@ -5002,9 +5186,12 @@ function createSkillsMemorySource(config = {}) {
 						}
 						case "run-start":
 							value = await service.startRun({
+								...request.signal === void 0 ? {} : { signal: request.signal },
 								skillName: memoryInputText(fields.skillName, "skillName", 64),
 								target: runTarget(fields.target),
-								caseIds: memoryInputStringArray(fields.caseIds, "caseIds", 20) ?? []
+								caseIds: memoryInputStringArray(fields.caseIds, "caseIds", 40) ?? [],
+								...fields.expectedCasesHash === void 0 ? {} : { expectedCasesHash: memoryInputText(fields.expectedCasesHash, "expectedCasesHash", 100) },
+								...backtestModel(fields.model)
 							});
 							break;
 						case "run-link":
@@ -5012,13 +5199,23 @@ function createSkillsMemorySource(config = {}) {
 							break;
 						case "run-progress":
 							value = await service.progressRun({
+								...request.signal === void 0 ? {} : { signal: request.signal },
 								id: memoryInputText(fields.id, "id", 100),
 								result: caseResult(fields.result)
 							});
 							break;
+						case "run-cancel":
+							value = await service.finishRun({
+								...request.signal === void 0 ? {} : { signal: request.signal },
+								id: memoryInputText(fields.id, "id", 100),
+								status: "interrupted",
+								error: "Backtest cancelled by the user."
+							});
+							break;
 						case "run-finish": {
-							const status = fields.status === "completed" ? "completed" : "failed";
+							const status = fields.status === "completed" ? "completed" : fields.status === "interrupted" ? "interrupted" : "failed";
 							value = await service.finishRun({
+								...request.signal === void 0 ? {} : { signal: request.signal },
 								id: memoryInputText(fields.id, "id", 100),
 								status,
 								...typeof fields.error === "string" ? { error: fields.error } : {}

package/lib/types/auto-adoption.d.ts

@@ -0,0 +1,9 @@
+import type { BacktestRun, EvalCase } from './contracts.ts';
+/** Stable evaluation inputs, excluding presentation timestamps and provenance. Hash at the server boundary. */
+export declare function skillCaseSignature(cases: readonly EvalCase[]): string;
+/** Automatic decisions require full matching coverage, not the UI's scored intersection. */
+export declare function compareAutomaticSkillRuns(cases: readonly EvalCase[], baseline: BacktestRun | undefined, draft: BacktestRun | undefined, casesHash: string): {
+    outcome: 'unmeasured' | 'regressed' | 'not-worse';
+    baselineScore?: number;
+    draftScore?: number;
+};

package/lib/types/client/api.d.ts

@@ -101,6 +101,7 @@ export interface SkillsPageClient {
     discardDraft(id: string): Promise<SkillDraft>;
     /** Attach a baseline run to the draft it is compared with. */
     linkRun(draftId: string, runId: string): Promise<void>;
+    cancelRun(id: string): Promise<BacktestRun>;
     /** The model each piece of model work follows: the two global picks and their per-item overrides. */
     draftPreferences(): Promise<DraftPreferencesView>;
     /** Pick the model for drafts, for backtests, or for one suggestion or skill; `null` follows the task Agent again. */
@@ -140,6 +141,7 @@ export interface SkillsPageClient {
         skill?: string;
         guidance?: string;
         targetRoot?: SkillRootKind;
+        backtest?: boolean;
         provider?: string;
         model?: string;
     }): Promise<{

package/lib/types/contracts.d.ts

@@ -174,6 +174,7 @@ export interface SkillSuggestion {
     status: SuggestionStatus;
     draftId?: string;
     lastError?: string;
+    autoAdoption?: AutoAdoptionEvent[];
     createdAt: string;
     updatedAt: string;
     expiresAt: string;
@@ -274,6 +275,8 @@ export interface CaseResult {
     error?: string;
     /** Kept under verdict=error so older readers also exclude it from scores. */
     unmeasuredReason?: 'requires-vision';
+    /** Every expected judge check was explicitly returned; absent on older runs. */
+    fullyGraded?: boolean;
 }
 export type RunStatus = 'running' | 'completed' | 'failed' | 'interrupted';
 export interface BacktestRun {
@@ -286,6 +289,11 @@ export interface BacktestRun {
         total: number;
     };
     caseIds: string[];
+    /** Source-verified case inputs at start; absent on older runs. */
+    casesHash?: string;
+    model?: ModelRoute & {
+        reasoningEffort?: string;
+    };
     results: CaseResult[];
     /** Over the graded cases; absent when none could be graded. */
     passRate?: number;
@@ -312,6 +320,10 @@ export interface ReviewLogEntry {
     error?: string;
     /** The conversation turn the reviewed window ended with. */
     turn?: number;
+    autoAdoptions?: Array<AutoAdoptionEvent & {
+        suggestionId: string;
+        skillName: string;
+    }>;
     /** The suggestions this review created or strengthened. */
     suggestionIds?: string[];
 }
@@ -415,6 +427,8 @@ export interface ReviewRecordResult {
     reviewId: string;
     trajectoryId?: string;
     suggestionIds: string[];
+    /** Newly inserted ids only; older Sources omit this capability. */
+    createdSuggestionIds?: string[];
     inserted: number;
     updated: number;
     casesAdded: number;
@@ -469,6 +483,8 @@ export interface SuggestionView {
 }
 export interface DraftView {
     draft: SkillDraft;
+    /** File changes need a new draft, never a retry with a newer revision. */
+    conflict?: 'base-changed' | 'target-exists';
     /** The file the draft replaces, as it is now. */
     base?: SkillDocument;
     resolution?: SkillResolution;
@@ -620,3 +636,26 @@ export interface MarketInstall {
     files: number;
     installedAt: string;
 }
+export declare const AUTO_ADOPTION_STAGES: readonly ["queued", "drafting", "backtesting", "rewriting", "adopted", "paused", "review"];
+export declare const AUTO_ADOPTION_REASONS: readonly ["disabled", "no-cases", "case-limit", "unmeasured", "regressed", "changed", "failed", "queue-full", "model-unavailable", "interrupted"];
+export interface AutoAdoptionEvent {
+    at: string;
+    stage: typeof AUTO_ADOPTION_STAGES[number];
+    attempt: number;
+    reason?: typeof AUTO_ADOPTION_REASONS[number];
+    detail?: string;
+    baselineScore?: number;
+    draftScore?: number;
+}
+export type DraftExpectation = {
+    id: string;
+    hash: string;
+} | null;
+export interface AutomaticAdoption {
+    attempt: number;
+    draftHash: string;
+    baselineRunId: string;
+    draftRunId: string;
+    casesHash: string;
+}
+export { skillCaseSignature, compareAutomaticSkillRuns } from './auto-adoption.ts';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-mnemon-source-skills",
-  "version": "0.5.9",
+  "version": "0.5.10",
   "description": "Skills Source for Composable View Memory: DSH skill management, evolution suggestions, evidence trajectories and eval sets.",
   "license": "MIT",
   "type": "module",
@@ -62,7 +62,7 @@
     "@types/node": "^22.0.0",
     "@types/react": "^18.3.0",
     "@types/react-dom": "^18.3.0",
-    "dsh-mnemon": "^0.5.42",
+    "dsh-mnemon": "^0.5.43",
     "jsdom": "^26.0.0",
     "react": "^18.2.0",
     "react-dom": "^18.2.0",

package/presentation/locales.json

@@ -353,7 +353,34 @@
     "evolution.loadFailedBody": "请重试，或在左侧选择其他建议。",
     "tests.model": "全局模型",
     "tests.modelCard": "所选回测模型",
-    "tests.modelAuto": "跟随默认"
+    "tests.modelAuto": "跟随默认",
+    "draft.baseChanged": "原技能已变更，此草案已过期。请基于当前文件重新生成并回测。",
+    "draft.targetExists": "同名技能已存在。请重新生成为改进草案并回测，再确认采纳。",
+    "draft.refreshGuidance": "基于当前技能文件重新生成，保留最近的修改，并重新运行回测。",
+    "draft.regenerate": "重新生成并回测",
+    "draft.cancelBacktest": "取消回测",
+    "draft.interrupted": "回测已中断。可以重新回测，或在检查草案后手动采纳。",
+    "evolution.auto.title": "自动采纳",
+    "evolution.auto.history": "自动采纳活动记录",
+    "evolution.auto.attempt": "第 {count} 轮",
+    "evolution.auto.scores": "基线 {baseline} / 草案 {draft}",
+    "evolution.auto.stage.queued": "等待处理",
+    "evolution.auto.stage.drafting": "正在起草",
+    "evolution.auto.stage.backtesting": "正在对比回测",
+    "evolution.auto.stage.rewriting": "依据退步结果重写",
+    "evolution.auto.stage.adopted": "已自动采纳",
+    "evolution.auto.stage.paused": "已停止自动处理",
+    "evolution.auto.stage.review": "需要人工审查",
+    "evolution.auto.reason.disabled": "自动采纳或所需权限已关闭",
+    "evolution.auto.reason.no-cases": "没有启用的回测用例",
+    "evolution.auto.reason.case-limit": "启用的用例超过每轮上限；调整用例或上限后手动处理本建议",
+    "evolution.auto.reason.unmeasured": "回测未完整测出所有检查项，或存在图片依赖、错误、中断",
+    "evolution.auto.reason.regressed": "至少一个用例或原先通过的检查项出现退步",
+    "evolution.auto.reason.changed": "建议、草案或用例已变动",
+    "evolution.auto.reason.failed": "自动处理失败，保留人工处理入口",
+    "evolution.auto.reason.queue-full": "自动处理队列已满",
+    "evolution.auto.reason.model-unavailable": "无法确定起草或回测模型",
+    "evolution.auto.reason.interrupted": "Host 已停止，不自动重启此建议"
   },
   "en": {
     "page.title": "Skills",
@@ -709,6 +736,33 @@
     "evolution.loadFailedBody": "Retry, or select another suggestion on the left.",
     "tests.model": "Global model",
     "tests.modelCard": "Selected model",
-    "tests.modelAuto": "Follow default"
+    "tests.modelAuto": "Follow default",
+    "draft.baseChanged": "The original skill changed. This draft is out of date; regenerate it against the current file and backtest it.",
+    "draft.targetExists": "A skill with this name now exists. Regenerate an update and backtest it before adopting.",
+    "draft.refreshGuidance": "Regenerate against the current skill file, preserving its latest changes, then run backtests.",
+    "draft.regenerate": "Regenerate and backtest",
+    "draft.cancelBacktest": "Cancel backtests",
+    "draft.interrupted": "Backtests were interrupted. Run them again, or review the draft before adopting manually.",
+    "evolution.auto.title": "Automatic adoption",
+    "evolution.auto.history": "Automatic adoption activity",
+    "evolution.auto.attempt": "Attempt {count}",
+    "evolution.auto.scores": "Baseline {baseline} / draft {draft}",
+    "evolution.auto.stage.queued": "Queued",
+    "evolution.auto.stage.drafting": "Drafting",
+    "evolution.auto.stage.backtesting": "Backtesting a pair",
+    "evolution.auto.stage.rewriting": "Rewriting after regression",
+    "evolution.auto.stage.adopted": "Automatically adopted",
+    "evolution.auto.stage.paused": "Automatic work stopped",
+    "evolution.auto.stage.review": "Manual review required",
+    "evolution.auto.reason.disabled": "Automatic adoption or its required permissions were disabled",
+    "evolution.auto.reason.no-cases": "No enabled evaluation cases",
+    "evolution.auto.reason.case-limit": "Enabled cases exceed the per-run limit; adjust cases or the limit and review this suggestion manually",
+    "evolution.auto.reason.unmeasured": "Some checks were not fully measured, or a case needs images, failed or was interrupted",
+    "evolution.auto.reason.regressed": "At least one case or previously passing check regressed",
+    "evolution.auto.reason.changed": "The suggestion, draft or evaluation cases changed",
+    "evolution.auto.reason.failed": "Automatic work failed; manual review remains available",
+    "evolution.auto.reason.queue-full": "Automatic queue is full",
+    "evolution.auto.reason.model-unavailable": "No resolved draft or backtest model",
+    "evolution.auto.reason.interrupted": "Host stopped; this suggestion will not restart automatically"
   }
 }