patchwarden 0.6.4 → 1.1.0

package/README.en.md

@@ -8,10 +8,9 @@
 [![Node.js](https://img.shields.io/badge/Node.js-%3E%3D18-339933.svg)](https://nodejs.org/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
 
-Current stable release: **v0.6.4**. See the
-[release notes](docs/release-v0.6.4.md),
-[migration guide](docs/migration-from-safe-bifrost.md), and
-[GitHub Release](https://github.com/jiezeng2004-design/PatchWarden/releases/tag/v0.6.4).
+Current source version: **v1.1.0**. See the
+[CHANGELOG](CHANGELOG.md), [migration guide](docs/migration-from-safe-bifrost.md), and
+[release checklist](docs/release-checklist.md). Verify GitHub Release / npm publication separately before release.
 
 PatchWarden is a local-first MCP safety and verification layer for AI coding
 agents, with workspace confinement, command allowlists, scope-violation
@@ -266,7 +265,7 @@ execution still requires a separate Watcher.
 New-Item -ItemType Directory .\patchwarden-runtime
 Set-Location .\patchwarden-runtime
 npm.cmd init -y
-npm.cmd install patchwarden@0.6.4
+npm.cmd install patchwarden@<published-version>
 Copy-Item .\node_modules\patchwarden\examples\config.example.json .\patchwarden.config.json
 $env:PATCHWARDEN_CONFIG = (Resolve-Path .\patchwarden.config.json)
 node .\node_modules\patchwarden\dist\runner\watch.js
@@ -275,10 +274,10 @@ node .\node_modules\patchwarden\dist\runner\watch.js
 An MCP client can launch:
 
 ```text
-npx.cmd -y patchwarden@0.6.4
+npx.cmd -y patchwarden@<published-version>
 ```
 
-Pin the version in important environments instead of using `latest`
+Replace `<published-version>` with a version that exists on npm/GitHub, and pin that version in important environments instead of using `latest`
 unconditionally.
 
 ## Complete configuration guide
@@ -331,7 +330,7 @@ Configuration fields:
 | `workspaceRoot` | Yes | The only workspace root PatchWarden may access. |
 | `plansDir` | Yes | Plan directory, normally `.patchwarden/plans`. |
 | `tasksDir` | Yes | Task and result directory, normally `.patchwarden/tasks`. |
-| `toolProfile` | No | `full` or `chatgpt_core`; use `full` for local clients. |
+| `toolProfile` | No | `full`, `chatgpt_core`, `chatgpt_direct`, or `chatgpt_search`; use `full` for local clients and `chatgpt_search` for compact discovery-driven clients. |
 | `agents` | Yes | Execution-agent allowlist; supports `{repo}` and `{prompt}` placeholders. |
 | `allowedTestCommands` | Yes | Exact allowlist for independent verification commands. |
 | `repoAllowedTestCommands` | No | Extra exact commands keyed by workspace-relative repository path; wildcards are unsupported. |
@@ -434,7 +433,7 @@ Pinned npm configuration:
 ```toml
 [mcp_servers.patchwarden]
 command = "npx.cmd"
-args = ["-y", "patchwarden@0.6.4"]
+args = ["-y", "patchwarden@<published-version>"]
 
 [mcp_servers.patchwarden.env]
 PATCHWARDEN_CONFIG = "D:\\path\\to\\patchwarden.config.json"
@@ -494,7 +493,7 @@ PatchWarden.cmd start core
 The launcher:
 
 - Builds `dist/index.js` if it is missing.
-- Verifies v0.6.4, the fixed 17-tool `chatgpt_core` catalog, and its schema
+- Verifies v1.1.0, the fixed 21-tool `chatgpt_core` catalog, and its schema
   manifest.
 - Reads or prompts for the Tunnel ID.
 - Reads or prompts for the runtime API key.
@@ -742,7 +741,7 @@ Optional token configuration:
 Set the token in the PowerShell process that starts the server:
 
 ```powershell
-$env:PATCHWARDEN_OWNER_TOKEN = "use-a-random-local-only-value"
+$env:PATCHWARDEN_OWNER_TOKEN = "<token>"
 ```
 
 Clients can send `Authorization: Bearer ...` or `x-patchwarden-token`. Never
@@ -931,13 +930,13 @@ names.
 
 ## MCP tools and profiles
 
-`chatgpt_core` is the fixed 17-tool profile used by the ChatGPT tunnel:
+`chatgpt_core` is the fixed 21-tool profile used by the ChatGPT tunnel:
 
 `health_check`, `list_agents`, `list_workspace`,
 `read_workspace_file`, `save_plan`, `create_task`,
 `wait_for_task`, `get_task_summary`, `get_diff`, `get_result`,
 `get_result_json`, `get_test_log`, `get_task_status`, `list_tasks`,
-`cancel_task`, `audit_task`, and `safe_status`.
+`cancel_task`, `audit_task`, `safe_status`, `safe_result`, `safe_audit`, `safe_test_summary`, and `safe_diff_summary`.
 
 `get_task_summary` keeps the backward-compatible `standard` view by default.
 ChatGPT should request `view: "compact"` first; terminal `wait_for_task`
@@ -957,7 +956,7 @@ to `full`.
 
 ### ChatGPT Direct mode
 
-Direct mode exposes ten guarded tools so ChatGPT can create an editing
+Direct mode exposes thirteen guarded tools so ChatGPT can create an editing
 session, read and search source files, apply hash-bound JSON patches, run
 exactly allowlisted verification commands, finalize the evidence, and audit
 the result without a local execution agent.
@@ -981,7 +980,7 @@ On first use, provide the `tunnel-client.exe` path and a Tunnel ID dedicated
 to the Direct Connector. The launcher uses the `patchwarden-direct` profile,
 stores runtime state under `%LOCALAPPDATA%\patchwarden\runtime-direct`, skips
 the Watcher, and retains the existing DPAPI credential handling. In a fresh
-ChatGPT conversation, `health_check` should report `chatgpt_direct`, ten
+ChatGPT conversation, `health_check` should report `chatgpt_direct`, thirteen
 tools, and `direct_profile_enabled=true`.
 
 ## Security boundaries and local data
@@ -1017,7 +1016,7 @@ Start with a dedicated test workspace and a repository you can recover.
 Upgrade a pinned npm installation:
 
 ```powershell
-npm.cmd install patchwarden@0.6.4
+npm.cmd install patchwarden@<published-version>
 ```
 
 Upgrade a source checkout:
@@ -1101,8 +1100,9 @@ and release-asset checksums independently.
 - [x] ChatGPT Connector / tunnel
 - [x] Doctor and runtime health checks
 - [x] Tool manifest and schema-drift detection
-- [ ] Worktree isolation
-- [ ] Multi-agent task queue
+- [x] Release Gate (five-stage pre-release verification)
+- [x] Worktree isolation
+- [x] Multi-agent routing
 - [ ] Local dashboard
 
 ## License

package/README.md

@@ -8,10 +8,9 @@
 [![Node.js](https://img.shields.io/badge/Node.js-%3E%3D18-339933.svg)](https://nodejs.org/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
 
-当前稳定版本：**v0.6.4**。查看
-[发布说明](docs/release-v0.6.4.md)、
-[迁移指南](docs/migration-from-safe-bifrost.md)和
-[GitHub Release](https://github.com/jiezeng2004-design/PatchWarden/releases/tag/v0.6.4)。
+当前源码版本：**v1.1.0**。查看
+[CHANGELOG](CHANGELOG.md)、[迁移指南](docs/migration-from-safe-bifrost.md)和
+[发布检查清单](docs/release-checklist.md)。GitHub Release / npm 发布状态需要在发布前单独核对。
 
 PatchWarden 是一个面向本地编程 Agent 的安全 MCP 桥接器。上游的
 ChatGPT、Codex、OpenCode 或其他 MCP 客户端负责规划与验收，
@@ -180,7 +179,7 @@ npm 包适合让 MCP 客户端通过固定版本启动 PatchWarden。为了执
 New-Item -ItemType Directory .\patchwarden-runtime
 Set-Location .\patchwarden-runtime
 npm.cmd init -y
-npm.cmd install patchwarden@0.6.4
+npm.cmd install patchwarden@<published-version>
 Copy-Item .\node_modules\patchwarden\examples\config.example.json .\patchwarden.config.json
 $env:PATCHWARDEN_CONFIG = (Resolve-Path .\patchwarden.config.json)
 node .\node_modules\patchwarden\dist\runner\watch.js
@@ -189,10 +188,10 @@ node .\node_modules\patchwarden\dist\runner\watch.js
 MCP 客户端可以使用：
 
 ```text
-npx.cmd -y patchwarden@0.6.4
+npx.cmd -y patchwarden@<published-version>
 ```
 
-建议固定版本号，不要在重要环境中无条件使用 `latest`。
+将 `<published-version>` 替换为已经在 npm/GitHub 上确认存在的版本；不要在重要环境中无条件使用 `latest`。
 
 ## 完整配置说明
 
@@ -244,7 +243,7 @@ Copy-Item .\examples\config.example.json .\patchwarden.config.json
 | `workspaceRoot` | 是 | PatchWarden 唯一允许访问的工作区根目录。 |
 | `plansDir` | 是 | 计划目录，通常使用 `.patchwarden/plans`。 |
 | `tasksDir` | 是 | 任务和结果目录，通常使用 `.patchwarden/tasks`。 |
-| `toolProfile` | 否 | `full` 或 `chatgpt_core`；本地客户端推荐 `full`。 |
+| `toolProfile` | 否 | `full`、`chatgpt_core`、`chatgpt_direct` 或 `chatgpt_search`；本地客户端推荐 `full`，动态工具发现场景使用 `chatgpt_search`。 |
 | `agents` | 是 | 可执行 Agent 白名单；支持 `{repo}` 和 `{prompt}` 占位符。 |
 | `allowedTestCommands` | 是 | 独立验证命令白名单，调用时必须精确匹配。 |
 | `repoAllowedTestCommands` | 否 | 按工作区相对仓库路径增加精确验证命令；不支持通配符。 |
@@ -342,7 +341,7 @@ npm.cmd run watch
 ```toml
 [mcp_servers.patchwarden]
 command = "npx.cmd"
-args = ["-y", "patchwarden@0.6.4"]
+args = ["-y", "patchwarden@<published-version>"]
 
 [mcp_servers.patchwarden.env]
 PATCHWARDEN_CONFIG = "D:\\path\\to\\patchwarden.config.json"
@@ -400,7 +399,7 @@ PatchWarden.cmd start core
 启动器会：
 
 - 检查 `dist/index.js`，缺失时自动构建。
-- 校验 `chatgpt_core` 的版本、16 个核心工具和 Schema Manifest。
+- 校验 `chatgpt_core` 的版本、21 个核心工具和 Schema Manifest。
 - 读取或提示输入 Tunnel ID。
 - 读取或提示输入运行时 API Key。
 - 使用 Windows DPAPI 保存凭据到 `%APPDATA%\patchwarden`。
@@ -837,18 +836,18 @@ PatchWarden v0.4.0 不自动读取旧 CLI、旧环境变量、旧 Header、旧
 
 ## MCP 工具与 Profile
 
-`chatgpt_core` 是固定的 17 工具 Profile，适合 ChatGPT Tunnel：
+`chatgpt_core` 是固定的 21 工具 Profile，适合 ChatGPT Tunnel：
 
 `health_check`、`list_agents`、`list_workspace`、
 `read_workspace_file`、`save_plan`、`create_task`、
 `wait_for_task`、`get_task_summary`、`get_diff`、`get_result`、
 `get_result_json`、`get_test_log`、`get_task_status`、`list_tasks`、
-`cancel_task`、`audit_task`、`safe_status`。
+`cancel_task`、`audit_task`、`safe_status`、`safe_result`、`safe_audit`、`safe_test_summary`、`safe_diff_summary`。
 
 `get_task_summary` 默认保留兼容的 `standard` 视图；ChatGPT 应优先使用
 `view: "compact"`，终态 `wait_for_task` 也只内嵌 compact 验收证据。
 
-`full` 提供完整 30 工具本地开发目录，包含核心工具、管理工具和 Direct
+`full` 提供完整 54 工具本地开发目录，包含核心工具、管理工具和 Direct
 工具。除 `chatgpt_core` 外，常用额外管理工具包括：
 
 - `get_plan`
@@ -858,15 +857,15 @@ PatchWarden v0.4.0 不自动读取旧 CLI、旧环境变量、旧 Header、旧
 - `get_task_stdout_tail`
 - `get_task_log_tail`
 
-`chatgpt_direct` 是 v0.6.0 新增的 Direct 直接开发 Profile，v0.6.4 包含 10 个工具：
+`chatgpt_direct` 是 v0.6.0 新增的 Direct 直接开发 Profile，v1.1 包含 13 个工具：
 
 `health_check`、`list_workspace`、`create_direct_session`、
 `search_workspace`、`read_workspace_file`、`apply_patch`、
-`run_verification`、`finalize_direct_session`、`audit_session`、`sync_file`。
+`run_verification`、`finalize_direct_session`、`audit_session`、`safe_direct_summary`、`safe_finalize_direct_session`、`safe_audit_direct_session`、`sync_file`。
 
 `chatgpt_direct` 默认关闭，需要通过 `enableDirectProfile: true` 或
 `PATCHWARDEN_TOOL_PROFILE=chatgpt_direct` 显式启用。`chatgpt_core` 保持
-17 工具不变。
+21 工具清单。
 
 Tunnel 包装脚本会强制使用 `chatgpt_core`；普通本地开发默认使用 `full`。
 
@@ -921,7 +920,7 @@ Tunnel ID。启动器使用 `patchwarden-direct` Profile、独立的
 Tunnel API Key 仍通过现有 Windows DPAPI 缓存处理，不会写入仓库。
 
 连接 ChatGPT 后新建对话并先调用 `health_check`。预期
-`tool_profile=chatgpt_direct`、`tool_count=10`、
+`tool_profile=chatgpt_direct`、`tool_count=13`、
 `direct_profile_enabled=true`。已有对话可能缓存旧工具清单，需要重新连接
 Connector 后再新建对话。
 
@@ -984,7 +983,7 @@ PatchWarden 能降低误操作风险，但不能替代人工审查。第一次
 升级 npm 固定版本：
 
 ```powershell
-npm.cmd install patchwarden@0.6.4
+npm.cmd install patchwarden@<published-version>
 ```
 
 源码升级：
@@ -1067,8 +1066,9 @@ Release 和发布资产校验值。
 - [x] ChatGPT Connector / Tunnel
 - [x] Doctor 与运行时健康检查
 - [x] Tool Manifest 与 Schema 漂移检测
-- [ ] Worktree 隔离
-- [ ] 多 Agent 任务队列
+- [x] Release Gate（发布前五阶段校验）
+- [x] Worktree 隔离
+- [x] 多 Agent 路由
 - [ ] 本地 Dashboard
 
 ## License

package/dist/agents/agentRouter.d.ts

@@ -0,0 +1,40 @@
+export interface AgentRouteInput {
+    goal?: string;
+    scope?: string[];
+    inline_plan?: string;
+    template?: string;
+    configuredAgents: string[];
+}
+export interface AgentRouteResult {
+    recommended_agent: string;
+    reason: string;
+    fallback: boolean;
+}
+export declare const AGENT_ROUTING_RULES: {
+    /** 大规模代码修改 → opencode */
+    readonly largeScope: {
+        readonly agent: "opencode";
+        readonly threshold: 10;
+    };
+    /** 单文件修改 → patchwarden-direct */
+    readonly singleFile: {
+        readonly agent: "patchwarden-direct";
+    };
+    /** 验收审计 → patchwarden-audit */
+    readonly audit: {
+        readonly agent: "patchwarden-audit";
+        readonly keywords: readonly ["审计", "audit", "验收", "verify"];
+    };
+    /** 复杂推理和重构 → codex */
+    readonly refactor: {
+        readonly agent: "codex";
+        readonly keywords: readonly ["重构", "refactor", "跨模块", "redesign"];
+    };
+    /** 文档整理 → claude（fallback 到 opencode） */
+    readonly documentation: {
+        readonly agent: "claude";
+        readonly fallbackAgent: "opencode";
+        readonly keywords: readonly ["文档", "readme", "changelog", "doc"];
+    };
+};
+export declare function routeAgent(input: AgentRouteInput): AgentRouteResult;

package/dist/agents/agentRouter.js

@@ -0,0 +1,95 @@
+import { PatchWardenError } from "../errors.js";
+// ── Routing rules (roadmap 9.3) ───────────────────────────────────
+export const AGENT_ROUTING_RULES = {
+    /** 大规模代码修改 → opencode */
+    largeScope: {
+        agent: "opencode",
+        threshold: 10, // scope.length > threshold
+    },
+    /** 单文件修改 → patchwarden-direct */
+    singleFile: {
+        agent: "patchwarden-direct",
+    },
+    /** 验收审计 → patchwarden-audit */
+    audit: {
+        agent: "patchwarden-audit",
+        keywords: ["审计", "audit", "验收", "verify"],
+    },
+    /** 复杂推理和重构 → codex */
+    refactor: {
+        agent: "codex",
+        keywords: ["重构", "refactor", "跨模块", "redesign"],
+    },
+    /** 文档整理 → claude（fallback 到 opencode） */
+    documentation: {
+        agent: "claude",
+        fallbackAgent: "opencode",
+        keywords: ["文档", "readme", "changelog", "doc"],
+    },
+};
+// ── Helpers ───────────────────────────────────────────────────────
+function containsAny(text, keywords) {
+    return keywords.some((keyword) => text.includes(keyword));
+}
+function resolveRoute(preferred, reason, configuredAgents) {
+    if (configuredAgents.includes(preferred)) {
+        return { recommended_agent: preferred, reason, fallback: false };
+    }
+    const fallback = configuredAgents[0];
+    return {
+        recommended_agent: fallback,
+        reason: `fallback: ${preferred} not configured, using ${fallback}`,
+        fallback: true,
+    };
+}
+// ── Route agent ───────────────────────────────────────────────────
+export function routeAgent(input) {
+    if (!input.configuredAgents || input.configuredAgents.length === 0) {
+        throw new PatchWardenError("no_agent_configured", "No agent is configured for routing.", "Configure at least one agent in patchwarden.config.json under the agents field.");
+    }
+    const scopeCount = input.scope?.length ?? 0;
+    const text = `${input.goal || ""} ${input.inline_plan || ""}`.toLowerCase();
+    // Rule 1: 大规模代码修改 → opencode (scope files > 10)
+    if (scopeCount > AGENT_ROUTING_RULES.largeScope.threshold) {
+        return resolveRoute(AGENT_ROUTING_RULES.largeScope.agent, `large scope (${scopeCount} files)`, input.configuredAgents);
+    }
+    // Rule 2: 单文件修改 → patchwarden-direct (scope files === 1)
+    if (scopeCount === 1) {
+        return resolveRoute(AGENT_ROUTING_RULES.singleFile.agent, "single file scope", input.configuredAgents);
+    }
+    // Rule 3: 验收审计 → patchwarden-audit
+    if (containsAny(text, AGENT_ROUTING_RULES.audit.keywords)) {
+        return resolveRoute(AGENT_ROUTING_RULES.audit.agent, "audit keywords", input.configuredAgents);
+    }
+    // Rule 4: 复杂推理和重构 → codex
+    if (containsAny(text, AGENT_ROUTING_RULES.refactor.keywords)) {
+        return resolveRoute(AGENT_ROUTING_RULES.refactor.agent, "refactor keywords", input.configuredAgents);
+    }
+    // Rule 5: 文档整理 → claude（fallback 到 opencode）
+    if (containsAny(text, AGENT_ROUTING_RULES.documentation.keywords)) {
+        const preferred = AGENT_ROUTING_RULES.documentation.agent;
+        const fallbackAgent = AGENT_ROUTING_RULES.documentation.fallbackAgent;
+        if (input.configuredAgents.includes(preferred)) {
+            return { recommended_agent: preferred, reason: "documentation keywords", fallback: false };
+        }
+        if (input.configuredAgents.includes(fallbackAgent)) {
+            return {
+                recommended_agent: fallbackAgent,
+                reason: `fallback: ${preferred} not configured, using ${fallbackAgent}`,
+                fallback: true,
+            };
+        }
+        const firstAgent = input.configuredAgents[0];
+        return {
+            recommended_agent: firstAgent,
+            reason: `fallback: ${preferred} not configured, using ${firstAgent}`,
+            fallback: true,
+        };
+    }
+    // Rule 6: 无匹配 → fallback 到 configuredAgents[0]
+    return {
+        recommended_agent: input.configuredAgents[0],
+        reason: `no specific routing rule matched, using default agent ${input.configuredAgents[0]}`,
+        fallback: true,
+    };
+}

package/dist/config.d.ts

@@ -15,7 +15,7 @@ export interface PatchWardenConfig {
     defaultTaskTimeoutSeconds: number;
     maxTaskTimeoutSeconds: number;
     watcherStaleSeconds: number;
-    toolProfile?: "full" | "chatgpt_core" | "chatgpt_direct";
+    toolProfile?: "full" | "chatgpt_core" | "chatgpt_direct" | "chatgpt_search";
     repoAliases?: Record<string, string>;
     enableAgentAssessment?: boolean;
     agentAssessmentTimeoutSeconds?: number;

package/dist/config.js

@@ -175,8 +175,12 @@ function normalizeConfig(config) {
     if (!Number.isInteger(config.watcherStaleSeconds) || config.watcherStaleSeconds < 5 || config.watcherStaleSeconds > 3600) {
         throw new Error("watcherStaleSeconds must be an integer from 5 to 3600");
     }
-    if (config.toolProfile !== undefined && config.toolProfile !== "full" && config.toolProfile !== "chatgpt_core" && config.toolProfile !== "chatgpt_direct") {
-        throw new Error('toolProfile must be "full", "chatgpt_core", or "chatgpt_direct"');
+    if (config.toolProfile !== undefined &&
+        config.toolProfile !== "full" &&
+        config.toolProfile !== "chatgpt_core" &&
+        config.toolProfile !== "chatgpt_direct" &&
+        config.toolProfile !== "chatgpt_search") {
+        throw new Error('toolProfile must be "full", "chatgpt_core", "chatgpt_direct", or "chatgpt_search"');
     }
     if (config.enableAgentAssessment !== undefined && typeof config.enableAgentAssessment !== "boolean") {
         throw new Error("enableAgentAssessment must be a boolean");

package/dist/controlCenter.js

@@ -419,6 +419,7 @@ function resolveWorkspaceRootSafe() {
 }
 const TERMINAL_TASK_STATUSES = new Set([
     "done",
+    "done_by_agent",
     "failed",
     "failed_verification",
     "failed_scope_violation",
@@ -928,13 +929,19 @@ function handleReconcile(res, taskId) {
         const statusData = readJsonFileSafe(statusPath) ?? {};
         const runtimeData = readJsonFileSafe(runtimePath) ?? {};
         const watcher = readWatcherStatusSafe();
+        const VALID_ACCEPTANCE = ["pending", "accepted", "rejected", "needs_fix", "blocked"];
+        const taskStatus = String(statusData.status || "pending");
+        const taskAcceptanceStatus = taskStatus === "done_by_agent"
+            ? (typeof statusData.acceptance_status === "string" && VALID_ACCEPTANCE.includes(statusData.acceptance_status) ? statusData.acceptance_status : "pending")
+            : null;
         const taskEntry = {
             task_id: taskId,
             plan_id: String(statusData.plan_id || ""),
             title: "",
             agent: String(statusData.agent || ""),
-            status: String(statusData.status || "pending"),
+            status: taskStatus,
             phase: String(runtimeData.phase || statusData.phase || "queued"),
+            acceptance_status: taskAcceptanceStatus,
             created_at: String(statusData.created_at || ""),
             updated_at: String(statusData.updated_at || ""),
             workspace_root: String(statusData.workspace_root || config.workspaceRoot),
@@ -1072,13 +1079,19 @@ function handleTaskDetail(res, taskId) {
         let stale = null;
         if (statusData) {
             const watcher = readWatcherStatusSafe();
+            const VALID_ACCEPTANCE2 = ["pending", "accepted", "rejected", "needs_fix", "blocked"];
+            const taskStatus2 = String(statusData.status || "pending");
+            const taskAcceptanceStatus2 = taskStatus2 === "done_by_agent"
+                ? (typeof statusData.acceptance_status === "string" && VALID_ACCEPTANCE2.includes(statusData.acceptance_status) ? statusData.acceptance_status : "pending")
+                : null;
             const entry = {
                 task_id: taskId,
                 plan_id: String(statusData.plan_id || ""),
                 title: "",
                 agent: String(statusData.agent || ""),
-                status: String(statusData.status || "pending"),
+                status: taskStatus2,
                 phase: String(runtimeData?.phase || statusData.phase || "queued"),
+                acceptance_status: taskAcceptanceStatus2,
                 created_at: String(statusData.created_at || ""),
                 updated_at: String(statusData.updated_at || ""),
                 workspace_root: String(statusData.workspace_root || config.workspaceRoot),

package/dist/doctor.js

@@ -18,6 +18,9 @@ import { guardPlanContent } from "./security/planGuard.js";
 import { TASK_READ_ONLY_FILES } from "./tools/getTaskFile.js";
 import { getToolDefs } from "./tools/registry.js";
 import { CHATGPT_CORE_TOOL_NAMES, CHATGPT_DIRECT_TOOL_NAMES, selectToolsForProfile } from "./tools/toolCatalog.js";
+import { buildToolRegistry } from "./tools/toolRegistry.js";
+import { runAllSchemaDriftChecks } from "./tools/schemaDriftCheck.js";
+import { runReleaseGateCheck } from "./release/releaseGate.js";
 import { PATCHWARDEN_VERSION } from "./version.js";
 // ── State ──────────────────────────────────────────────────────────
 let ok = 0;
@@ -218,7 +221,7 @@ async function main() {
         const coreTools = selectToolsForProfile(fullTools, "chatgpt_core", config?.enableDirectProfile);
         const createSchema = coreTools.find((tool) => tool.name === "create_task")?.inputSchema;
         const waitSchema = coreTools.find((tool) => tool.name === "wait_for_task")?.inputSchema;
-        check("Full tool profile exposes 30 tools", fullTools.length === 30, `${fullTools.length} tools`);
+        check("Full tool profile exposes 54 tools", fullTools.length === 54, `${fullTools.length} tools`);
         check(`chatgpt_core profile exposes the exact ${CHATGPT_CORE_TOOL_NAMES.length}-tool manifest`, JSON.stringify(coreTools.map((tool) => tool.name)) === JSON.stringify(CHATGPT_CORE_TOOL_NAMES), coreTools.map((tool) => tool.name).join(", "));
         check("Core task schemas expose inline_plan, verify_commands, and wait aliases", Boolean(createSchema?.properties?.inline_plan &&
             createSchema?.properties?.verify_commands &&
@@ -235,6 +238,25 @@ async function main() {
             results.push(`[OK]   chatgpt_direct enabled check skipped (enableDirectProfile=false)`);
             ok++;
         }
+        // 16. Schema drift 检查（v0.9.0）—— warn 级别，不阻断 doctor:ci
+        // 用 fullTools 同时构建 registry 和 toolDefs Map，确保 schema digest 比对基准一致。
+        const driftRegistry = buildToolRegistry(fullTools);
+        const driftToolDefs = new Map();
+        for (const tool of fullTools) {
+            driftToolDefs.set(tool.name, { inputSchema: tool.inputSchema });
+        }
+        const driftResult = runAllSchemaDriftChecks(driftRegistry, driftToolDefs);
+        if (driftResult.ok) {
+            results.push(`[OK]   Schema drift check — no drift detected`);
+            ok++;
+        }
+        else {
+            // 每个 warning 单独输出为 WARN，与其他自检项的多警告模式一致
+            for (const w of driftResult.warnings) {
+                results.push(`[WARN] Schema drift: ${w}`);
+                warn++;
+            }
+        }
     }
     finally {
         if (previousProfile === undefined)
@@ -242,6 +264,15 @@ async function main() {
         else
             process.env.PATCHWARDEN_TOOL_PROFILE = previousProfile;
     }
+    // 17. Release gate module loadable (v1.0.0) — module integrity only.
+    // Does NOT execute local_ready (would recurse into doctor:ci).
+    try {
+        const releaseGateReady = typeof runReleaseGateCheck === "function";
+        check("Release gate module loadable", releaseGateReady, releaseGateReady ? "runReleaseGateCheck exported" : "runReleaseGateCheck missing or invalid");
+    }
+    catch (error) {
+        check("Release gate module loadable", false, error instanceof Error ? error.message : String(error));
+    }
     // 9. HTTP port check
     const httpPort = config?.http?.port || 7331;
     try {

package/dist/goal/acceptanceEngine.d.ts

@@ -0,0 +1,62 @@
+/**
+ * v0.7.2: 验收引擎 — 基于证据推导 4 级验收结论。
+ *
+ * 输入：status.json、task_meta（goal/scope/forbidden/verification/done_evidence）、
+ *       audit checks（pass/warn/fail）、release claims、artifact 信息
+ * 输出：ACCEPTED / NEEDS_FIX / REJECTED / BLOCKED_BY_APPROVAL
+ *       + reasons + required_evidence + next_suggested_task
+ *
+ * 决策规则：
+ *   1. 存在 fail 级检查项 → REJECTED
+ *   2. 存在 release claims 但无法验证远端 → BLOCKED_BY_APPROVAL
+ *   3. 存在 warn 级检查项或文档不同步 → NEEDS_FIX
+ *   4. 所有检查通过 → ACCEPTED
+ */
+import type { AcceptanceStatus } from "../tools/createTask.js";
+export type AcceptanceVerdict = "ACCEPTED" | "NEEDS_FIX" | "REJECTED" | "BLOCKED_BY_APPROVAL";
+export interface AcceptanceCheck {
+    name: string;
+    result: "pass" | "warn" | "fail";
+    detail: string;
+}
+export interface AcceptanceEvidence {
+    task_id: string;
+    task_status: string;
+    result_md_exists: boolean;
+    result_json_exists: boolean;
+    verify_json_exists: boolean;
+    test_log_exists: boolean;
+    git_diff_exists: boolean;
+    verify_status: "passed" | "failed" | "skipped" | null;
+    new_out_of_scope_changes: Array<{
+        path: string;
+        change: string;
+    }>;
+    goal: string | null;
+    scope: string[] | null;
+    forbidden: string[] | null;
+    verification: string[] | null;
+    done_evidence: string[] | null;
+    artifact_status: string | null;
+    release_claims_unverified: boolean;
+    checks: AcceptanceCheck[];
+}
+export interface AcceptanceResult {
+    verdict: AcceptanceVerdict;
+    acceptance_status: AcceptanceStatus;
+    reason: string;
+    reasons: string[];
+    required_evidence: string[];
+    next_suggested_task: string;
+    fail_checks: AcceptanceCheck[];
+    warn_checks: AcceptanceCheck[];
+}
+/**
+ * 核心验收决策函数。
+ * 输入所有证据，输出 4 级验收结论。
+ */
+export declare function evaluateAcceptance(evidence: AcceptanceEvidence): AcceptanceResult;
+/**
+ * 将 AcceptanceVerdict 映射到 AcceptanceStatus（用于回写 status.json）。
+ */
+export declare function verdictToStatus(verdict: AcceptanceVerdict): AcceptanceStatus;