@alfe.ai/gateway 0.10.8 → 0.11.0

package/dist/health.js

@@ -7,7 +7,7 @@ import { chmodSync, existsSync, mkdirSync, readFileSync, unlinkSync, writeFileSy
 import { ClaudeCodeApplier, ClaudeCodeMcpSync, HermesApplier, HermesMcpSync, IntegrationManager, IntegrationManagerAdapter, LockManager, McpApplier, NoopOpenClawCliLock, OpenClawApplier, SerialOpenClawCliLock } from "@alfe.ai/integrations";
 import { execFile, execSync } from "node:child_process";
 import { promisify } from "node:util";
-import { getEndpointFromToken, readConfig } from "@alfe.ai/config";
+import { getEndpointFromToken, readConfig, resolveAgentWorkspace } from "@alfe.ai/config";
 import { AgentApiClient } from "@alfe.ai/agent-api-client";
 import { parse } from "smol-toml";
 import WebSocket from "ws";
@@ -3744,11 +3744,23 @@ function superRefine(fn) {
 * new host, then cut a CLI release so the pin travels with the CLI version. The
 * dashboard still surfaces npm-`latest` as an explicit opt-in upgrade.
 *
-* NOTE: claude-code is self-hosted-only (no managed Docker image), so — unlike
+* NOTE: managed claude-code agents run on Hetzner VMs provisioned by cloud-init
+* + `alfe setup --managed` (no Docker image), so — unlike
 * `PINNED_OPENCLAW_VERSION` — there is no `services/compute/Dockerfile` mirror to
 * keep in sync.
 */
-const PINNED_CLAUDE_CODE_HOST_VERSION = "0.1.6";
+const PINNED_CLAUDE_CODE_HOST_VERSION = "0.1.15";
+/**
+* The `@anthropic-ai/claude-code` (the `claude` CLI) version Alfe installs for
+* the claude-code runtime: always on managed VMs (`alfe setup --managed`), and
+* on self-hosted machines only when no `claude` is present (a user-installed
+* `claude` is never replaced). Pinned to npm's `stable` dist-tag as of
+* 2026-10-14 — `latest` moves daily and the host drives `claude -p` flags
+* (`--resume`, `--mcp-config`, `--strict-mcp-config`, `--permission-mode`)
+* whose behaviour must be validated before a bump. The pin travels with the
+* CLI release.
+*/
+const PINNED_CLAUDE_CODE_VERSION = "2.1.287";
 //#endregion
 //#region ../../packages-internal/ids/dist/prefixes.js
 const ID_PREFIXES = {
@@ -4388,13 +4400,18 @@ const AnthropicModel = {
 	Opus48: "claude-opus-4-8",
 	Opus47: "claude-opus-4-7",
 	Opus46: "claude-opus-4-6",
+	Sonnet55: "claude-sonnet-5-5",
 	Sonnet5: "claude-sonnet-5",
 	Sonnet46: "claude-sonnet-4-6",
+	Haiku55: "claude-haiku-5-5",
 	Haiku45: "claude-haiku-4-5"
 };
 const ANTHROPIC_MODELS = enumValues(AnthropicModel);
 const OpenAIModel = {
 	GPT6Astra: "gpt-6-astra",
+	GPT61Sol: "gpt-6.1-sol",
+	GPT6Sol: "gpt-6-sol",
+	GPT6Luna: "gpt-6-luna",
 	GPT4o: "gpt-4o",
 	GPT4oMini: "gpt-4o-mini",
 	O3: "o3",
@@ -4424,8 +4441,8 @@ const DeepSeekModel = {
 const DEEPSEEK_MODELS = enumValues(DeepSeekModel);
 const GoogleModel = {
 	Gemini38Flash: "gemini-3.8-flash",
-	Gemini37Flash: "gemini-3.7-flash",
 	Gemini36Flash: "gemini-3.6-flash",
+	Gemini37Flash: "gemini-3.7-flash",
 	Gemini35Flash: "gemini-3.5-flash",
 	Gemini35FlashLite: "gemini-3.5-flash-lite",
 	Gemini31Pro: "gemini-3.1-pro-preview",
@@ -4439,12 +4456,11 @@ const MiniMaxModel = {
 	M3: "MiniMax-M3",
 	M27: "MiniMax-M2.7",
 	M27HighSpeed: "MiniMax-M2.7-highspeed",
-	M25: "MiniMax-M2.5",
-	M21: "MiniMax-M2.1",
-	M2: "MiniMax-M2"
+	M25: "MiniMax-M2.5"
 };
 const MINIMAX_MODELS = enumValues(MiniMaxModel);
 const MistralModel = {
+	Large4: "mistral-large-4",
 	Large: "mistral-large-latest",
 	Medium: "mistral-medium-latest",
 	Small: "mistral-small-latest",
@@ -4514,7 +4530,7 @@ _enum(ZHIPU_MODELS);
 _enum(MOONSHOT_MODELS);
 _enum(QWEN_MODELS);
 string().trim().min(1);
-AnthropicModel.Opus55, AnthropicModel.Opus5, AnthropicModel.Opus48, AnthropicModel.Opus47, AnthropicModel.Opus46, AnthropicModel.Sonnet5, AnthropicModel.Sonnet46, AnthropicModel.Haiku45, OpenAIModel.GPT6Astra, OpenAIModel.GPT4o, OpenAIModel.GPT4oMini, OpenAIModel.O3, OpenAIModel.GPT41, OpenAIModel.GPT41Mini, OpenAIModel.GPT41Nano, OpenAIModel.GPT54, OpenAIModel.GPT54Mini, OpenAIModel.GPT54Nano, OpenAIModel.GPT54Pro, OpenAIModel.GPT55, OpenAIModel.GPT55Pro, OpenAIModel.GPT56Sol, OpenAIModel.GPT56Terra, OpenAIModel.GPT56Luna, OpenAIModel.O3Mini, OpenAIModel.O4Mini, OpenAIModel.TextEmbedding3Small, OpenAIModel.TextEmbedding3Large, DeepSeekModel.Flash, DeepSeekModel.V4Pro, DeepSeekModel.V4Flash, GoogleModel.Gemini38Flash, GoogleModel.Gemini37Flash, GoogleModel.Gemini36Flash, GoogleModel.Gemini35Flash, GoogleModel.Gemini35FlashLite, GoogleModel.Gemini31Pro, GoogleModel.Gemini31FlashLite, GoogleModel.Gemini25Pro, GoogleModel.Gemini25Flash, GoogleModel.Gemini25FlashLite, MiniMaxModel.M3, MiniMaxModel.M27, MiniMaxModel.M27HighSpeed, MiniMaxModel.M25, MiniMaxModel.M21, MiniMaxModel.M2, MistralModel.Large, MistralModel.Medium, MistralModel.Small, MistralModel.Codestral, MistralModel.Ministral8b, MistralModel.Ministral3b, XAIModel.Grok47, XAIModel.Grok46, XAIModel.Grok45, XAIModel.Grok43, XAIModel.GrokBuild01, XAIModel.Grok4, XAIModel.Grok41Fast, ZhipuModel.GLM53, ZhipuModel.GLM53Flash, ZhipuModel.GLM53FlashX, ZhipuModel.GLM52, ZhipuModel.GLM51, ZhipuModel.GLM46, ZhipuModel.GLM45, ZhipuModel.GLM45Air, MoonshotModel.K3, MoonshotModel.K26, MoonshotModel.K27Code, MoonshotModel.K27CodeHighSpeed, QwenModel.Qwen38Max, QwenModel.Qwen38Flash, QwenModel.Qwen37Max, QwenModel.Qwen37Plus, QwenModel.Qwen36Flash, QwenModel.Qwen35Flash;
+AnthropicModel.Opus55, AnthropicModel.Opus5, AnthropicModel.Opus48, AnthropicModel.Opus47, AnthropicModel.Opus46, AnthropicModel.Sonnet55, AnthropicModel.Sonnet5, AnthropicModel.Sonnet46, AnthropicModel.Haiku55, AnthropicModel.Haiku45, OpenAIModel.GPT6Astra, OpenAIModel.GPT61Sol, OpenAIModel.GPT6Sol, OpenAIModel.GPT6Luna, OpenAIModel.GPT4o, OpenAIModel.GPT4oMini, OpenAIModel.O3, OpenAIModel.GPT41, OpenAIModel.GPT41Mini, OpenAIModel.GPT41Nano, OpenAIModel.GPT54, OpenAIModel.GPT54Mini, OpenAIModel.GPT54Nano, OpenAIModel.GPT54Pro, OpenAIModel.GPT55, OpenAIModel.GPT55Pro, OpenAIModel.GPT56Sol, OpenAIModel.GPT56Terra, OpenAIModel.GPT56Luna, OpenAIModel.O3Mini, OpenAIModel.O4Mini, OpenAIModel.TextEmbedding3Small, OpenAIModel.TextEmbedding3Large, DeepSeekModel.Flash, DeepSeekModel.V4Pro, DeepSeekModel.V4Flash, GoogleModel.Gemini38Flash, GoogleModel.Gemini37Flash, GoogleModel.Gemini36Flash, GoogleModel.Gemini35Flash, GoogleModel.Gemini35FlashLite, GoogleModel.Gemini31Pro, GoogleModel.Gemini31FlashLite, GoogleModel.Gemini25Pro, GoogleModel.Gemini25Flash, GoogleModel.Gemini25FlashLite, MiniMaxModel.M3, MiniMaxModel.M27, MiniMaxModel.M27HighSpeed, MiniMaxModel.M25, MistralModel.Large4, MistralModel.Large, MistralModel.Medium, MistralModel.Small, MistralModel.Codestral, MistralModel.Ministral8b, MistralModel.Ministral3b, XAIModel.Grok47, XAIModel.Grok46, XAIModel.Grok45, XAIModel.Grok43, XAIModel.GrokBuild01, XAIModel.Grok4, XAIModel.Grok41Fast, ZhipuModel.GLM53, ZhipuModel.GLM53Flash, ZhipuModel.GLM53FlashX, ZhipuModel.GLM52, ZhipuModel.GLM51, ZhipuModel.GLM46, ZhipuModel.GLM45, ZhipuModel.GLM45Air, MoonshotModel.K3, MoonshotModel.K26, MoonshotModel.K27Code, MoonshotModel.K27CodeHighSpeed, QwenModel.Qwen38Max, QwenModel.Qwen38Flash, QwenModel.Qwen37Max, QwenModel.Qwen37Plus, QwenModel.Qwen36Flash, QwenModel.Qwen35Flash;
 /**
 * Alfe-hosted open-weight models are promoted dynamically only in regions
 * whose private data plane exists. Keep them out of ALL_MODELS so the standard
@@ -4574,7 +4590,7 @@ const AWS_PRIVATE_MODEL_PRICING = {
 	[BedrockModel.TitanEmbedTextV2]: { input: .026004 }
 };
 Object.fromEntries(AWS_PRIVATE_BEDROCK_CHAT_MODELS.map((modelId) => [`alfe-bedrock-${modelId}`, AWS_PRIVATE_MODEL_PRICING[modelId]]));
-AnthropicModel.Opus55, AnthropicModel.Opus5, AnthropicModel.Opus48, AnthropicModel.Opus47, AnthropicModel.Opus46, AnthropicModel.Sonnet5, AnthropicModel.Sonnet46, AnthropicModel.Haiku45, OpenAIModel.GPT6Astra, OpenAIModel.GPT4o, OpenAIModel.GPT4oMini, OpenAIModel.O3, OpenAIModel.GPT41, OpenAIModel.GPT41Mini, OpenAIModel.GPT41Nano, OpenAIModel.GPT54, OpenAIModel.GPT54Mini, OpenAIModel.GPT54Nano, OpenAIModel.GPT54Pro, OpenAIModel.GPT55, OpenAIModel.GPT55Pro, OpenAIModel.GPT56Sol, OpenAIModel.GPT56Terra, OpenAIModel.GPT56Luna, OpenAIModel.O3Mini, OpenAIModel.O4Mini, OpenAIModel.TextEmbedding3Small, OpenAIModel.TextEmbedding3Large, DeepSeekModel.Flash, DeepSeekModel.V4Pro, DeepSeekModel.V4Flash, GoogleModel.Gemini38Flash, GoogleModel.Gemini37Flash, GoogleModel.Gemini36Flash, GoogleModel.Gemini35Flash, GoogleModel.Gemini35FlashLite, GoogleModel.Gemini31Pro, GoogleModel.Gemini31FlashLite, GoogleModel.Gemini25Pro, GoogleModel.Gemini25Flash, GoogleModel.Gemini25FlashLite, MiniMaxModel.M3, MiniMaxModel.M27, MiniMaxModel.M27HighSpeed, MiniMaxModel.M25, MiniMaxModel.M21, MiniMaxModel.M2, MistralModel.Large, MistralModel.Medium, MistralModel.Small, MistralModel.Codestral, MistralModel.Ministral8b, MistralModel.Ministral3b, XAIModel.Grok47, XAIModel.Grok46, XAIModel.Grok45, XAIModel.Grok43, XAIModel.GrokBuild01, XAIModel.Grok4, XAIModel.Grok41Fast, ZhipuModel.GLM53, ZhipuModel.GLM53Flash, ZhipuModel.GLM53FlashX, ZhipuModel.GLM52, ZhipuModel.GLM51, ZhipuModel.GLM46, ZhipuModel.GLM45, ZhipuModel.GLM45Air, MoonshotModel.K3, MoonshotModel.K26, MoonshotModel.K27Code, MoonshotModel.K27CodeHighSpeed, QwenModel.Qwen38Max, QwenModel.Qwen38Flash, QwenModel.Qwen37Max, QwenModel.Qwen37Plus, QwenModel.Qwen36Flash, QwenModel.Qwen35Flash;
+AnthropicModel.Opus55, AnthropicModel.Opus5, AnthropicModel.Opus48, AnthropicModel.Opus47, AnthropicModel.Opus46, AnthropicModel.Sonnet55, AnthropicModel.Sonnet5, AnthropicModel.Sonnet46, AnthropicModel.Haiku55, AnthropicModel.Haiku45, OpenAIModel.GPT6Astra, OpenAIModel.GPT61Sol, OpenAIModel.GPT6Sol, OpenAIModel.GPT6Luna, OpenAIModel.GPT4o, OpenAIModel.GPT4oMini, OpenAIModel.O3, OpenAIModel.GPT41, OpenAIModel.GPT41Mini, OpenAIModel.GPT41Nano, OpenAIModel.GPT54, OpenAIModel.GPT54Mini, OpenAIModel.GPT54Nano, OpenAIModel.GPT54Pro, OpenAIModel.GPT55, OpenAIModel.GPT55Pro, OpenAIModel.GPT56Sol, OpenAIModel.GPT56Terra, OpenAIModel.GPT56Luna, OpenAIModel.O3Mini, OpenAIModel.O4Mini, OpenAIModel.TextEmbedding3Small, OpenAIModel.TextEmbedding3Large, DeepSeekModel.Flash, DeepSeekModel.V4Pro, DeepSeekModel.V4Flash, GoogleModel.Gemini38Flash, GoogleModel.Gemini37Flash, GoogleModel.Gemini36Flash, GoogleModel.Gemini35Flash, GoogleModel.Gemini35FlashLite, GoogleModel.Gemini31Pro, GoogleModel.Gemini31FlashLite, GoogleModel.Gemini25Pro, GoogleModel.Gemini25Flash, GoogleModel.Gemini25FlashLite, MiniMaxModel.M3, MiniMaxModel.M27, MiniMaxModel.M27HighSpeed, MiniMaxModel.M25, MistralModel.Large4, MistralModel.Large, MistralModel.Medium, MistralModel.Small, MistralModel.Codestral, MistralModel.Ministral8b, MistralModel.Ministral3b, XAIModel.Grok47, XAIModel.Grok46, XAIModel.Grok45, XAIModel.Grok43, XAIModel.GrokBuild01, XAIModel.Grok4, XAIModel.Grok41Fast, ZhipuModel.GLM53, ZhipuModel.GLM53Flash, ZhipuModel.GLM53FlashX, ZhipuModel.GLM52, ZhipuModel.GLM51, ZhipuModel.GLM46, ZhipuModel.GLM45, ZhipuModel.GLM45Air, MoonshotModel.K3, MoonshotModel.K26, MoonshotModel.K27Code, MoonshotModel.K27CodeHighSpeed, QwenModel.Qwen38Max, QwenModel.Qwen38Flash, QwenModel.Qwen37Max, QwenModel.Qwen37Plus, QwenModel.Qwen36Flash, QwenModel.Qwen35Flash;
 enumValues({
 	Standard: "standard",
 	AwsPrivate: "aws_private"
@@ -4718,6 +4734,14 @@ NotificationType.PaymentSucceeded, NotificationCategory.Billing, RecipientStrate
 * Platform-agnostic: works in React Native and browser environments.
 * Uses standard fetch() API — no Node.js dependencies.
 */
+/** A sane 429 window hint: a positive whole number of seconds (≤ 1 day). */
+function isRetryAfterSeconds(value) {
+	return typeof value === "number" && Number.isInteger(value) && value > 0 && value <= 86400;
+}
+/** Narrow allowlist: a non-empty list of `{ profile: string, message: string }`. */
+function isProfileErrorList(value) {
+	return Array.isArray(value) && value.length > 0 && value.every((entry) => typeof entry === "object" && entry !== null && typeof entry.profile === "string" && typeof entry.message === "string");
+}
 var AlfeApiClient = class {
 	apiBaseUrl;
 	getToken;
@@ -4790,7 +4814,12 @@ var AlfeApiClient = class {
 						error: errorMessage ?? (typeof body === "string" ? body : `API error: ${String(res.status)}`),
 						status: res.status,
 						...typeof errBody?.code === "string" ? { code: errBody.code } : {},
-						...Array.isArray(errBody?.ineligibleUserIds) && errBody.ineligibleUserIds.every((id) => typeof id === "string") ? { ineligibleUserIds: [...errBody.ineligibleUserIds] } : {}
+						...Array.isArray(errBody?.ineligibleUserIds) && errBody.ineligibleUserIds.every((id) => typeof id === "string") ? { ineligibleUserIds: [...errBody.ineligibleUserIds] } : {},
+						...isProfileErrorList(errBody?.errors) ? { profileErrors: errBody.errors.map(({ profile, message }) => ({
+							profile,
+							message
+						})) } : {},
+						...isRetryAfterSeconds(errBody?.retryAfterSeconds) ? { retryAfterSeconds: errBody.retryAfterSeconds } : {}
 					}
 				};
 			}
@@ -5517,16 +5546,13 @@ function _resolveManagedGatewayWsUrl(apiEndpoint) {
 *   workspace = "~/.openclaw"
 */
 /**
-* Derive the agent workspace path from the runtime's home dir.
-*
-* For openclaw, the agent CWD is `${home}/workspace` (matches OpenClaw's
-* `agents.defaults.workspace` default — see OpenClaw README "Agent workspace
-* + skills"). For other runtimes, default to home itself.
+* Derive the agent workspace path (the runtime's CWD) from its home dir.
+* Delegates to the shared `@alfe.ai/config` rule so the daemon, the CLI and
+* the claude-code host cannot drift: openclaw and claude-code run in
+* `${home}/workspace`, other runtimes run in the home itself.
 */
 function deriveAgentWorkspace(runtime, home) {
-	if (runtime === "openclaw") return join(home, "workspace");
-	if (runtime === "claude-code") return join(home, "workspace");
-	return home;
+	return resolveAgentWorkspace(runtime, home);
 }
 const UNSAFE_RECORD_KEYS = new Set([
 	"__proto__",
@@ -5589,6 +5615,7 @@ async function loadManagedConfig() {
 		runtime: identity.runtime,
 		gatewayWsUrl
 	}, "Managed config resolved");
+	const defaultModel = identity.runtime === "claude-code" ? await resolveClaudeCodeDefaultModel(apiKey, apiEndpoint) : void 0;
 	return {
 		apiKey,
 		apiEndpoint,
@@ -5599,7 +5626,8 @@ async function loadManagedConfig() {
 		orgId: identity.orgId,
 		runtime: identity.runtime,
 		runtimes: buildManagedRuntimes(identity.runtime),
-		autoStartRuntime: true
+		autoStartRuntime: true,
+		...defaultModel ? { defaultModel } : {}
 	};
 }
 /**
@@ -5622,6 +5650,13 @@ function buildManagedRuntimes(runtime) {
 			agentWorkspace: deriveAgentWorkspace("hermes", home)
 		} };
 	}
+	if (runtime === "claude-code") {
+		const home = join(homedir(), ".claude-code");
+		return { "claude-code": {
+			workspace: home,
+			agentWorkspace: deriveAgentWorkspace("claude-code", home)
+		} };
+	}
 	return {};
 }
 /**
@@ -5644,17 +5679,7 @@ async function loadDaemonConfig() {
 	const gatewayWsUrl = alfeConfig.gateway_url ?? deriveGatewayWsUrl(apiEndpoint);
 	const runtimes = await loadRuntimeConfigs();
 	const runtime = alfeConfig.runtime ?? identity.runtime;
-	let defaultModel;
-	if (runtime === "claude-code") {
-		let timer;
-		const ws = await Promise.race([fetchAgentConfig(alfeConfig.api_key, apiEndpoint), new Promise((resolve) => {
-			timer = setTimeout(() => {
-				resolve(null);
-			}, 5e3);
-		})]);
-		if (timer) clearTimeout(timer);
-		defaultModel = ws?.defaultModel;
-	}
+	const defaultModel = runtime === "claude-code" ? await resolveClaudeCodeDefaultModel(alfeConfig.api_key, apiEndpoint) : void 0;
 	return {
 		apiKey: alfeConfig.api_key,
 		apiEndpoint,
@@ -5670,6 +5695,38 @@ async function loadDaemonConfig() {
 	};
 }
 /**
+* The claude-code agent's bare `defaultModel`, for `ALFE_DEFAULT_MODEL`.
+* Bounded so a slow or hung workspace endpoint can't block daemon startup; any
+* failure or timeout yields `undefined` and the host falls back to the `claude`
+* CLI default. Shared by the managed and self-hosted config loaders.
+*/
+async function resolveClaudeCodeDefaultModel(apiKey, apiEndpoint, options = {}) {
+	const fetcher = options.fetcher ?? fetchWorkspaceDefaultModel;
+	let timer;
+	try {
+		return (await Promise.race([fetcher(apiKey, apiEndpoint), new Promise((resolve) => {
+			timer = setTimeout(() => {
+				resolve(null);
+			}, options.timeoutMs ?? 5e3);
+		})]))?.defaultModel ?? void 0;
+	} catch {
+		return;
+	} finally {
+		if (timer) clearTimeout(timer);
+	}
+}
+/**
+* Workspace metadata only. The default model lives on the workspace record, so
+* it must not wait on (or time out behind) the persona/template file fetch
+* that `fetchAgentConfig` also performs.
+*/
+async function fetchWorkspaceDefaultModel(apiKey, apiEndpoint) {
+	return { defaultModel: (await new AgentApiClient({
+		apiKey,
+		apiUrl: apiEndpoint
+	}).getWorkspace()).defaultModel ?? void 0 };
+}
+/**
 * Fetch agent workspace config from the API.
 *
 * 1. GET /agents/me/workspace → { templateKey }
@@ -7092,8 +7149,10 @@ var ConfigReconciler = class {
 	getApplier;
 	verifyRetries;
 	verifyRetryDelayMs;
+	onKeysApplied;
 	constructor(options) {
 		this.getApplier = options.getApplier;
+		this.onKeysApplied = options.onKeysApplied;
 		this.verifyRetries = options.verifyRetries ?? DEFAULT_VERIFY_RETRIES;
 		this.verifyRetryDelayMs = options.verifyRetryDelayMs ?? DEFAULT_VERIFY_RETRY_DELAY_MS;
 	}
@@ -7116,6 +7175,7 @@ var ConfigReconciler = class {
 		};
 		const failures = [];
 		const inconclusiveKeys = [];
+		const changedKeys = [];
 		for (const [key, want] of entries) try {
 			if (getConfigRaw) {
 				let current;
@@ -7136,8 +7196,11 @@ var ConfigReconciler = class {
 			if (getConfigRaw) {
 				const outcome = await this.verify(getConfigRaw, key, want);
 				if (outcome === "mismatched") failures.push(`${key}: verify failed (read-back did not match)`);
-				else if (outcome === "inconclusive") inconclusiveKeys.push(key);
-			}
+				else {
+					if (outcome === "inconclusive") inconclusiveKeys.push(key);
+					changedKeys.push(key);
+				}
+			} else changedKeys.push(key);
 		} catch (err) {
 			const msg = err instanceof Error ? err.message : String(err);
 			if (getConfigRaw) {
@@ -7147,15 +7210,18 @@ var ConfigReconciler = class {
 						key,
 						err: msg
 					}, "config set errored but value landed — treating as applied");
+					changedKeys.push(key);
 					continue;
 				}
 				if (outcome === "inconclusive") {
 					inconclusiveKeys.push(key);
+					changedKeys.push(key);
 					continue;
 				}
 			}
 			failures.push(`${key}: ${msg}`);
 		}
+		this.notifyKeysApplied(changedKeys);
 		if (failures.length > 0) {
 			const joined = failures.join("; ");
 			const reason = joined.length > 480 ? `${joined.slice(0, 480)}… (truncated)` : joined;
@@ -7185,6 +7251,17 @@ var ConfigReconciler = class {
 			status: "applied"
 		};
 	}
+	notifyKeysApplied(keys) {
+		if (keys.length === 0 || !this.onKeysApplied) return;
+		try {
+			this.onKeysApplied(keys);
+		} catch (err) {
+			log$4.warn({
+				keys,
+				err: err instanceof Error ? err.message : String(err)
+			}, "onKeysApplied hook failed — config report unaffected");
+		}
+	}
 	async verify(getConfigRaw, key, want) {
 		let observedValue = false;
 		for (let attempt = 0; attempt <= this.verifyRetries; attempt++) {
@@ -7203,6 +7280,38 @@ var ConfigReconciler = class {
 		return observedValue ? "mismatched" : "inconclusive";
 	}
 };
+/** Extra environment for the runtime child, layered over the daemon's own env. */
+function buildRuntimeEnv(input) {
+	if (input.runtime !== "claude-code") return {};
+	return {
+		...input.defaultModel ? { ALFE_DEFAULT_MODEL: input.defaultModel } : {},
+		...input.managed ? {
+			ALFE_MANAGED: "1",
+			DISABLE_AUTOUPDATER: "1"
+		} : {}
+	};
+}
+/**
+* True when a config reconcile that wrote `keys` must restart the runtime child.
+* Only the claude-code host needs it: it reads its credential once per start.
+* OpenClaw/Hermes hot-reload their state-shaped keys.
+*/
+function shouldRestartRuntimeForConfigKeys(runtime, keys) {
+	return runtime === "claude-code" && keys.includes("alfe.claudeCredential.revision");
+}
+/**
+* The `ConfigReconciler.onKeysApplied` hook the daemon installs: restart the
+* runtime (through the shared, turn-draining restart path) when a reconcile
+* wrote a new claude-code credential revision. The reconciler value-diffs, so
+* a reconnect re-sending the same revision does not restart.
+*/
+function createCredentialRestartHook(runtime, requestRestart, onRestart) {
+	return (keys) => {
+		if (!shouldRestartRuntimeForConfigKeys(runtime, keys)) return;
+		onRestart?.();
+		requestRestart();
+	};
+}
 //#endregion
 //#region src/command-queue.ts
 const DEFAULT_TTL_MS$1 = 300 * 1e3;
@@ -7483,7 +7592,8 @@ function generateLaunchdPlist() {
 * ~13,570 restarts over 2 days).
 *
 * The guard:
-*   - checks the CLI is intact: the symlink target (`dist/index.js`) exists AND
+*   - checks the CLI is intact: the symlink target (the package's `bin.alfe`,
+*     `dist/bin.js` today) exists AND
 *     `alfe --version` exits 0;
 *   - if broken, reinstalls `@alfe.ai/cli@$ALFE_CLI_VERSION` (the exact version
 *     baked into the unit's `Environment=`; falls back to `@latest` when
@@ -7507,7 +7617,7 @@ set -u
 TARGET='${exactVersion ? `@alfe.ai/cli@${exactVersion}` : "@alfe.ai/cli@latest"}'
 log() { echo "[alfe-cli-guard] $*" >&2; }
 
-# Resolve the alfe bin and its real target (dist/index.js). readlink -f follows
+# Resolve the alfe bin and its real target (dist/bin.js). readlink -f follows
 # the whole symlink chain; fall back to the bin path itself if unavailable.
 BIN="$(command -v alfe 2>/dev/null || true)"
 if [ -n "$BIN" ]; then
@@ -10250,12 +10360,17 @@ async function startDaemon() {
 	const runtimeGate = new RuntimeGate(() => runtimeProcess);
 	runtimeGateRef = runtimeGate;
 	cloudClient.setIntegrationManager(integrationAdapter, runtimeGate);
-	cloudClient.setConfigReconciler(new ConfigReconciler({ getApplier: () => runtimeAppliers.get(config.runtime) }));
 	const requestRuntimeRestart = () => {
 		logger.info("Runtime state changed — requesting runtime restart");
 		runtimeGate.requestRestart();
 	};
 	cloudClient.setRuntimeRestartNeededHandler(requestRuntimeRestart);
+	cloudClient.setConfigReconciler(new ConfigReconciler({
+		getApplier: () => runtimeAppliers.get(config.runtime),
+		onKeysApplied: createCredentialRestartHook(config.runtime, requestRuntimeRestart, () => {
+			logger.info({ runtime: config.runtime }, "Claude credential changed — restarting runtime to refetch");
+		})
+	}));
 	if (config.runtime === "hermes") {
 		const hermesCfg = config.runtimes[config.runtime];
 		if (hermesCfg) {
@@ -10324,7 +10439,11 @@ async function startDaemon() {
 			runtimeProcess = new RuntimeProcess({
 				runtime: config.runtime,
 				workspace: runtimeCfg.workspace,
-				env: config.runtime === "claude-code" && config.defaultModel ? { ALFE_DEFAULT_MODEL: config.defaultModel } : {}
+				env: buildRuntimeEnv({
+					runtime: config.runtime,
+					...config.defaultModel ? { defaultModel: config.defaultModel } : {},
+					managed
+				})
 			});
 			runtimeProcess.setTurnActivityProbe(turnActivityProbe);
 			if (firstRuntimeStart) {
@@ -10542,7 +10661,7 @@ async function executeCloudCommand(command) {
 				pinnedVersion: target.version
 			}, "runtime.update requested an OpenClaw version above the pin — upgrading to the pin instead");
 			version = target.version;
-		} else version = payload?.version ?? (runtime === "claude-code" ? "0.1.6" : void 0);
+		} else version = payload?.version ?? (runtime === "claude-code" ? "0.1.15" : void 0);
 		upgradingRuntime = true;
 		setTimeout(() => {
 			(async () => {
@@ -11017,4 +11136,4 @@ function formatDuration(ms) {
 	return `${String(Math.round(seconds / 3600))}h`;
 }
 //#endregion
-export { loadDaemonConfig as _, installService as a, startService as c, uninstallService as d, PROTOCOL_VERSION as f, fetchAgentConfig as g, SOCKET_PATH as h, checkExistingDaemon as i, stopExistingDaemon as l, PID_PATH as m, queryDaemonHealth as n, isServiceInstalled as o, ALFE_DIR as p, startDaemon as r, restartService as s, formatHealthReport as t, stopService as u, resolveAgentIdentity as v, PINNED_CLAUDE_CODE_HOST_VERSION as y };
+export { loadDaemonConfig as _, installService as a, PINNED_CLAUDE_CODE_VERSION as b, startService as c, uninstallService as d, PROTOCOL_VERSION as f, fetchAgentConfig as g, SOCKET_PATH as h, checkExistingDaemon as i, stopExistingDaemon as l, PID_PATH as m, queryDaemonHealth as n, isServiceInstalled as o, ALFE_DIR as p, startDaemon as r, restartService as s, formatHealthReport as t, stopService as u, resolveAgentIdentity as v, PINNED_CLAUDE_CODE_HOST_VERSION as y };

package/dist/src/index.d.ts

@@ -145,11 +145,23 @@ declare function remediateManagedNode(runCommand: RemediationCommandRunner, base
  * new host, then cut a CLI release so the pin travels with the CLI version. The
  * dashboard still surfaces npm-`latest` as an explicit opt-in upgrade.
  *
- * NOTE: claude-code is self-hosted-only (no managed Docker image), so — unlike
+ * NOTE: managed claude-code agents run on Hetzner VMs provisioned by cloud-init
+ * + `alfe setup --managed` (no Docker image), so — unlike
  * `PINNED_OPENCLAW_VERSION` — there is no `services/compute/Dockerfile` mirror to
  * keep in sync.
  */
-declare const PINNED_CLAUDE_CODE_HOST_VERSION = "0.1.6";
+declare const PINNED_CLAUDE_CODE_HOST_VERSION = "0.1.15";
+/**
+ * The `@anthropic-ai/claude-code` (the `claude` CLI) version Alfe installs for
+ * the claude-code runtime: always on managed VMs (`alfe setup --managed`), and
+ * on self-hosted machines only when no `claude` is present (a user-installed
+ * `claude` is never replaced). Pinned to npm's `stable` dist-tag as of
+ * 2026-10-14 — `latest` moves daily and the host drives `claude -p` flags
+ * (`--resume`, `--mcp-config`, `--strict-mcp-config`, `--permission-mode`)
+ * whose behaviour must be validated before a bump. The pin travels with the
+ * CLI release.
+ */
+declare const PINNED_CLAUDE_CODE_VERSION = "2.1.287";
 //#endregion
 //#region src/sentry.d.ts
 /**
@@ -391,6 +403,13 @@ declare function resolveAgentIdentity(apiKey: string, apiEndpoint: string): Prom
  * In managed mode, loads from environment variables instead.
  */
 declare function loadDaemonConfig(): Promise<DaemonConfig>;
+/**
+ * The claude-code agent's bare `defaultModel`, for `ALFE_DEFAULT_MODEL`.
+ * Bounded so a slow or hung workspace endpoint can't block daemon startup; any
+ * failure or timeout yields `undefined` and the host falls back to the `claude`
+ * CLI default. Shared by the managed and self-hosted config loaders.
+ */
+
 /**
  * Fetch agent workspace config from the API.
  *
@@ -495,4 +514,4 @@ declare function checkExistingDaemon(): Promise<number | null>;
  */
 declare function stopExistingDaemon(): Promise<boolean>;
 //#endregion
-export { AGENT_DAEMON_SENTRY_DSN, AGENT_MCP_SENTRY_DSN, AGENT_RUNTIME_SENTRY_DSN, ALFE_DIR, type AgentIdentity, type AgentWorkspaceConfig, type DaemonConfig, type DaemonHealth, type IPCEvent, type IPCRequest, type IPCResponse, type InitAgentSentryOptions, OPENCLAW_MANAGED_NODE_MAJOR, OPENCLAW_NODE_REQUIREMENT, PID_PATH, PINNED_CLAUDE_CODE_HOST_VERSION, PINNED_OPENCLAW_VERSION, PROTOCOL_VERSION, type RemediateManagedNodeOptions, type RemediationCommandRunner, type RuntimeOutputLine, SOCKET_PATH, type SentrySurface, captureCliFailure, captureFatal, captureIntegrationFailure, captureMcpFailure, captureRuntimeCrash, captureRuntimeErrorOutput, checkExistingDaemon, fetchAgentConfig, flushSentry, formatHealthReport, initAgentSentry, installService, isServiceInstalled, loadDaemonConfig, logger, queryDaemonHealth, remediateManagedNode, resolveAgentIdentity, restartService, setAgentContext, startDaemon, startService, stopExistingDaemon, stopService, supportsOpenClawNode, uninstallService };
+export { AGENT_DAEMON_SENTRY_DSN, AGENT_MCP_SENTRY_DSN, AGENT_RUNTIME_SENTRY_DSN, ALFE_DIR, type AgentIdentity, type AgentWorkspaceConfig, type DaemonConfig, type DaemonHealth, type IPCEvent, type IPCRequest, type IPCResponse, type InitAgentSentryOptions, OPENCLAW_MANAGED_NODE_MAJOR, OPENCLAW_NODE_REQUIREMENT, PID_PATH, PINNED_CLAUDE_CODE_HOST_VERSION, PINNED_CLAUDE_CODE_VERSION, PINNED_OPENCLAW_VERSION, PROTOCOL_VERSION, type RemediateManagedNodeOptions, type RemediationCommandRunner, type RuntimeOutputLine, SOCKET_PATH, type SentrySurface, captureCliFailure, captureFatal, captureIntegrationFailure, captureMcpFailure, captureRuntimeCrash, captureRuntimeErrorOutput, checkExistingDaemon, fetchAgentConfig, flushSentry, formatHealthReport, initAgentSentry, installService, isServiceInstalled, loadDaemonConfig, logger, queryDaemonHealth, remediateManagedNode, resolveAgentIdentity, restartService, setAgentContext, startDaemon, startService, stopExistingDaemon, stopService, supportsOpenClawNode, uninstallService };

package/dist/src/index.js

@@ -1,5 +1,5 @@
 import { b as PINNED_OPENCLAW_VERSION, v as OPENCLAW_MANAGED_NODE_MAJOR, w as supportsOpenClawNode, y as OPENCLAW_NODE_REQUIREMENT } from "../runtime-upgrade-recovery.js";
-import { _ as loadDaemonConfig, a as installService, c as startService, d as uninstallService, f as PROTOCOL_VERSION, g as fetchAgentConfig, h as SOCKET_PATH, i as checkExistingDaemon, l as stopExistingDaemon, m as PID_PATH, n as queryDaemonHealth, o as isServiceInstalled, p as ALFE_DIR, r as startDaemon, s as restartService, t as formatHealthReport, u as stopService, v as resolveAgentIdentity, y as PINNED_CLAUDE_CODE_HOST_VERSION } from "../health.js";
+import { _ as loadDaemonConfig, a as installService, b as PINNED_CLAUDE_CODE_VERSION, c as startService, d as uninstallService, f as PROTOCOL_VERSION, g as fetchAgentConfig, h as SOCKET_PATH, i as checkExistingDaemon, l as stopExistingDaemon, m as PID_PATH, n as queryDaemonHealth, o as isServiceInstalled, p as ALFE_DIR, r as startDaemon, s as restartService, t as formatHealthReport, u as stopService, v as resolveAgentIdentity, y as PINNED_CLAUDE_CODE_HOST_VERSION } from "../health.js";
 import { a as captureFatal, d as flushSentry, f as initAgentSentry, h as logger, i as captureCliFailure, l as captureRuntimeCrash, n as AGENT_MCP_SENTRY_DSN, o as captureIntegrationFailure, p as setAgentContext, r as AGENT_RUNTIME_SENTRY_DSN, s as captureMcpFailure, t as AGENT_DAEMON_SENTRY_DSN, u as captureRuntimeErrorOutput } from "../sentry.js";
 import { n as remediateManagedNode } from "../node-remediation.js";
-export { AGENT_DAEMON_SENTRY_DSN, AGENT_MCP_SENTRY_DSN, AGENT_RUNTIME_SENTRY_DSN, ALFE_DIR, OPENCLAW_MANAGED_NODE_MAJOR, OPENCLAW_NODE_REQUIREMENT, PID_PATH, PINNED_CLAUDE_CODE_HOST_VERSION, PINNED_OPENCLAW_VERSION, PROTOCOL_VERSION, SOCKET_PATH, captureCliFailure, captureFatal, captureIntegrationFailure, captureMcpFailure, captureRuntimeCrash, captureRuntimeErrorOutput, checkExistingDaemon, fetchAgentConfig, flushSentry, formatHealthReport, initAgentSentry, installService, isServiceInstalled, loadDaemonConfig, logger, queryDaemonHealth, remediateManagedNode, resolveAgentIdentity, restartService, setAgentContext, startDaemon, startService, stopExistingDaemon, stopService, supportsOpenClawNode, uninstallService };
+export { AGENT_DAEMON_SENTRY_DSN, AGENT_MCP_SENTRY_DSN, AGENT_RUNTIME_SENTRY_DSN, ALFE_DIR, OPENCLAW_MANAGED_NODE_MAJOR, OPENCLAW_NODE_REQUIREMENT, PID_PATH, PINNED_CLAUDE_CODE_HOST_VERSION, PINNED_CLAUDE_CODE_VERSION, PINNED_OPENCLAW_VERSION, PROTOCOL_VERSION, SOCKET_PATH, captureCliFailure, captureFatal, captureIntegrationFailure, captureMcpFailure, captureRuntimeCrash, captureRuntimeErrorOutput, checkExistingDaemon, fetchAgentConfig, flushSentry, formatHealthReport, initAgentSentry, installService, isServiceInstalled, loadDaemonConfig, logger, queryDaemonHealth, remediateManagedNode, resolveAgentIdentity, restartService, setAgentContext, startDaemon, startService, stopExistingDaemon, stopService, supportsOpenClawNode, uninstallService };

package/dist/upgrade.js

@@ -21,8 +21,8 @@ import { promisify } from "node:util";
 *      broken) `alfe` binary. Keep the two in lockstep.
 *
 * The two checks:
-*   - the resolved CLI entry (`.../dist/index.js`, the symlink target of the
-*     `alfe` bin) exists on disk; and
+*   - the resolved CLI entry (the package's `bin.alfe`, `.../dist/bin.js`
+*     today, i.e. the symlink target of the `alfe` bin) exists on disk; and
 *   - running it (`node <entry> --version`, or `alfe --version`) exits 0 and
 *     prints the expected version.
 *
@@ -61,7 +61,7 @@ function decideCliInstallOk(probe, expectedVersion) {
 /**
 * Resolve the CLI entry file this daemon is running from. When the gateway is
 * loaded as a dependency of the globally-installed CLI the layout is
-* `.../@alfe.ai/cli/dist/index.js` with `.../@alfe.ai/gateway/dist/...`
+* `.../@alfe.ai/cli/dist/bin.js` with `.../@alfe.ai/gateway/dist/...`
 * alongside — walk up to the sibling `@alfe.ai/cli/package.json` and derive its
 * `bin.alfe` / main entry. Returns undefined when it can't be resolved (e.g.
 * `alfe gateway daemon` run from a dev checkout); the caller then skips the
@@ -106,7 +106,7 @@ async function tryVersion(cmd, args) {
 * Probe + decide whether the on-disk global CLI install is intact.
 *
 * @param expectedVersion  when set, the reported version must match exactly.
-* @param entryPath        the resolved `.../dist/index.js` (from
+* @param entryPath        the resolved `bin.alfe` entry, `.../dist/bin.js` (from
 *                         `resolveCliEntryPath`). When provided and missing,
 *                         the check fails fast. Also used to run the CLI
 *                         directly via `node <entry> --version`, which is more
@@ -174,7 +174,7 @@ function findOrphanReifyTempDirs(siblings, packageBasename) {
 * Detection order (robust — never hardcode `/usr/lib`):
 *   1. `npm root -g` → the global `node_modules` root. Authoritative on the box.
 *   2. Derive from the resolved CLI entry path (`.../node_modules/@alfe.ai/cli/
-*      dist/index.js`) by walking up to the `node_modules` segment.
+*      dist/bin.js`) by walking up to the `node_modules` segment.
 *
 * Returns undefined when neither resolves (dev checkout / non-global run); the
 * caller then skips the sweep (there is no global install to clean).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@alfe.ai/gateway",
-  "version": "0.10.8",
+  "version": "0.11.0",
   "description": "Alfe local gateway daemon — persistent control plane for agent integrations",
   "type": "module",
   "bin": {
@@ -23,11 +23,11 @@
     "pino-roll": "^1.2.0",
     "smol-toml": ">=1.6.1",
     "ws": "^8.18.0",
-    "@alfe.ai/agent-api-client": "^0.19.2",
+    "@alfe.ai/agent-api-client": "^0.21.0",
     "@alfe.ai/ai-proxy-local": "^0.0.17",
-    "@alfe.ai/config": "^0.4.1",
-    "@alfe.ai/integration-manifest": "^0.4.2",
-    "@alfe.ai/integrations": "^0.6.15",
+    "@alfe.ai/config": "^0.4.2",
+    "@alfe.ai/integration-manifest": "^0.5.0",
+    "@alfe.ai/integrations": "^0.7.0",
     "@alfe.ai/mcp-bundler": "^0.4.3"
   },
   "license": "UNLICENSED",