@steerable/agent-harness 0.2.3 → 0.2.5

package/dist/generated/ActionSegmentPayload.d.ts

@@ -0,0 +1,21 @@
+/**
+ * Payload of an action-segment card -- a strip of tool / action invocations the agent ran inline within an assistant message. Each segment carries a kind (the action type), a status, and arbitrary args/output payload that downstream renderers pretty-print.
+ */
+export interface ActionSegmentPayload {
+    segments: {
+        id: string;
+        /**
+         * Action / tool name, e.g. 'create_task' or 'search.web'.
+         */
+        kind: string;
+        status: "pending" | "running" | "succeeded" | "failed" | "cancelled";
+        label?: string | null;
+        args?: any;
+        output?: any;
+        error?: string | null;
+        startedAt?: string | null;
+        finishedAt?: string | null;
+        [k: string]: any;
+    }[];
+    [k: string]: any;
+}

package/dist/generated/ActionSegmentPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/AnalysisDocumentPayload.d.ts

@@ -0,0 +1,19 @@
+/**
+ * Payload of an analysis-document card. Renders as a long-form markdown document (research write-up, design memo, etc.). The body string can include GFM markdown, mermaid blocks, and inline references. The optional metadata block carries model / timestamp info shown in the document header.
+ */
+export interface AnalysisDocumentPayload {
+    /**
+     * Optional document title; defaults to a generic 'Analysis' label client-side.
+     */
+    title?: string | null;
+    /**
+     * Markdown body. May include mermaid code-fences.
+     */
+    body: string;
+    createdAt?: string | null;
+    /**
+     * Display model badge id (resolved to model option client-side).
+     */
+    modelId?: string | null;
+    [k: string]: any;
+}

package/dist/generated/AnalysisDocumentPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/AskUserQuestionsPayload.d.ts

@@ -0,0 +1,20 @@
+/**
+ * Payload of an ask-user-questions card. The agent has paused mid-run and needs structured input from the user before continuing. Each question is one of: option buttons (select), free-form text, or password. After the user answers, the backend re-emits the same card with answers filled in to switch the UI to read-only.
+ */
+export interface AskUserQuestionsPayload {
+    intro: string;
+    outro?: string | null;
+    answers?: {
+        [k: string]: string | string[];
+    } | null;
+    questions: {
+        id: string;
+        text: string;
+        type?: "select" | "text" | "password";
+        options?: string[];
+        placeholder?: string | null;
+        multiSelect?: boolean;
+        [k: string]: any;
+    }[];
+    [k: string]: any;
+}

package/dist/generated/AskUserQuestionsPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/CoverageReportPayload.d.ts

@@ -0,0 +1,42 @@
+/**
+ * Payload of a coverage-report card. Summarises mastery of a knowledge graph: section-level coverage / mastery and a list of weak knowledge points the learner should remediate. The actions block tells the UI whether a 'practice weak points' button should be offered.
+ */
+export interface CoverageReportPayload {
+    reportId: string;
+    title: string;
+    /**
+     * 0..1 fraction of nodes covered.
+     */
+    overallCoverage: number;
+    /**
+     * 0..1 fraction of nodes mastered.
+     */
+    overallMastery: number;
+    summary?: string | null;
+    sections: {
+        id: string;
+        name: string;
+        coverage: number;
+        mastery: number;
+        totalCount: number;
+        learnedCount: number;
+        testedCount: number;
+        masteredCount: number;
+        weakKpIds: string[];
+        [k: string]: any;
+    }[];
+    weakPoints: {
+        id: string;
+        name: string;
+        sectionName?: string | null;
+        accuracy: number;
+        recommendation: string;
+        [k: string]: any;
+    }[];
+    actions: {
+        allowRemediateQuiz: boolean;
+        remediateActionLabel: string;
+        [k: string]: any;
+    };
+    [k: string]: any;
+}

package/dist/generated/CoverageReportPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/OrchestrationPlanPayload.d.ts

@@ -0,0 +1,43 @@
+/**
+ * Payload of an orchestration-plan card emitted by the Coordinator agent. Lists every worker task (id + agent + prompt + dependencies). The UI renders one row per task with a live status dot driven by sibling assistant messages tagged with the same orchestrationGroupId + orchestrationTaskId.
+ */
+export interface OrchestrationPlanPayload {
+    /**
+     * Short prose explanation of why the coordinator picked this plan shape.
+     */
+    rationale?: string;
+    /**
+     * How the workers run with respect to each other.
+     */
+    mode?: "parallel" | "sequential" | "dag";
+    tasks: {
+        /**
+         * Stable id for this task within the orchestration group.
+         */
+        id: string;
+        /**
+         * Which agent runs this task.
+         */
+        agentId: string;
+        /**
+         * Per-task prompt the coordinator drafted.
+         */
+        prompt?: string;
+        /**
+         * Task ids that must finish before this one starts.
+         */
+        dependsOn?: string[];
+        /**
+         * Task ids whose outputs this task is allowed to read.
+         */
+        readOutputsFrom?: string[];
+        [k: string]: any;
+    }[];
+    coordinator?: {
+        agentId?: string;
+        name?: string;
+        avatar?: string;
+        [k: string]: any;
+    };
+    [k: string]: any;
+}

package/dist/generated/OrchestrationPlanPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/PlanSelectorPayload.d.ts

@@ -0,0 +1,37 @@
+/**
+ * Payload of a plan-selector card. The agent has drafted multiple candidate plans and is asking the user to pick one. Each plan carries effort / risk metrics and a pros/cons list. After selection the backend re-emits with selectedPlan filled in.
+ */
+export interface PlanSelectorPayload {
+    /**
+     * Short prose summarising how the candidate plans differ.
+     */
+    comparison: string;
+    /**
+     * Id of the user-chosen plan once decided.
+     */
+    selectedPlan?: string | null;
+    goalAttribution: {
+        type: "existing" | "new";
+        existingGoalId?: string | null;
+        existingGoalTitle?: string | null;
+        newGoalTitle?: string | null;
+        [k: string]: any;
+    };
+    plans: {
+        id: string;
+        name: string;
+        summary: string;
+        approach: string;
+        bestFor: string;
+        metrics: {
+            duration: string;
+            effortLevel: "low" | "medium" | "high";
+            riskLevel: "low" | "medium" | "high";
+            [k: string]: any;
+        };
+        pros: string[];
+        cons: string[];
+        [k: string]: any;
+    }[];
+    [k: string]: any;
+}

package/dist/generated/PlanSelectorPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/PlanStepsPayload.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Payload of a plan-steps card -- a compact horizontal checklist of skill names the agent intends to execute. Rendered as 'step1 / step2 / step3 ...' inline.
+ */
+export interface PlanStepsPayload {
+    steps: string[];
+    [k: string]: any;
+}

package/dist/generated/PlanStepsPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/QuizPayload.d.ts

@@ -0,0 +1,28 @@
+/**
+ * Payload of a quiz card. Carries one or more questions (choice / fill / judge / short_answer) and a submit button label. After the learner submits, the backend re-emits the same quizId with submittedAnswers filled in to switch the UI to read-only review mode.
+ */
+export interface QuizPayload {
+    /**
+     * Stable id, used for resubmit and review continuity.
+     */
+    quizId: string;
+    title: string;
+    description?: string | null;
+    submitActionLabel: string;
+    submittedAnswers?: {
+        [k: string]: string | string[];
+    } | null;
+    questions: {
+        id: string;
+        type: "choice" | "fill" | "judge" | "short_answer";
+        stem: string;
+        options?: string[] | null;
+        allowMultiple?: boolean | null;
+        placeholder?: string | null;
+        points?: number | null;
+        knowledgePointId?: string | null;
+        difficulty?: number | null;
+        [k: string]: any;
+    }[];
+    [k: string]: any;
+}

package/dist/generated/QuizPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/ResearchPlanPayload.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Payload of a research-plan card. Snapshots a research agent's current sub-question tree and its decision for the next round. The UI shows each sub-question with kind / evidence-strength badges.
+ */
+export interface ResearchPlanPayload {
+    topic: string;
+    round: number;
+    /**
+     * True when the research loop is complete and this snapshot is the last one.
+     */
+    final: boolean;
+    subQuestions: {
+        id: string;
+        question: string;
+        kind: "fact" | "compare" | "conclusion" | "risk";
+        status: "pending" | "searching" | "evidenced_strong" | "evidenced_medium" | "evidenced_weak" | "conflicted" | "exhausted";
+        evidenceCount: number;
+        note?: string | null;
+        [k: string]: any;
+    }[];
+    decision: {
+        next: "continue" | "expand" | "converge";
+        reason?: string | null;
+        [k: string]: any;
+    };
+    [k: string]: any;
+}

package/dist/generated/ResearchPlanPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/SearchSourcesPayload.d.ts

@@ -0,0 +1,14 @@
+/**
+ * Payload of a search-sources card. Lists the web pages that an agent consulted while answering. Rendered as a row of stacked favicons with an expandable detail list.
+ */
+export interface SearchSourcesPayload {
+    sources: {
+        url: string;
+        title?: string | null;
+        snippet?: string | null;
+        favicon?: string | null;
+        publishedAt?: string | null;
+        [k: string]: any;
+    }[];
+    [k: string]: any;
+}

package/dist/generated/SearchSourcesPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/SuggestedRepliesPayload.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Payload of a suggested-replies card. A small set of single-shot quick-reply texts shown below the composer; clicking one sends that text as the next user message.
+ */
+export interface SuggestedRepliesPayload {
+    /**
+     * Quick-reply texts. Clients typically render the first ~6; backend should keep the list short.
+     */
+    suggestions: string[];
+    [k: string]: any;
+}

package/dist/generated/SuggestedRepliesPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/SummaryMessagePayload.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Payload of a summary-message card. Used when the chat history was condensed: the card shows a markdown summary along with a count of how many original messages it replaces. Collapsed by default.
+ */
+export interface SummaryMessagePayload {
+    /**
+     * Markdown summary text.
+     */
+    body: string;
+    summarizedCount?: number | null;
+    status?: "pending" | "complete" | "failed" | null;
+    /**
+     * Free-form summary kind label (e.g. 'history_compaction').
+     */
+    type?: string | null;
+    [k: string]: any;
+}

package/dist/generated/SummaryMessagePayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/ThinkingProcessPayload.d.ts

@@ -0,0 +1,11 @@
+/**
+ * Payload of a thinking-process card. Renders the agent's chain-of-thought (markdown) in a collapsible panel. UI clients are responsible for redacting / disabling this card if the deployment policy forbids exposing raw reasoning to end users.
+ */
+export interface ThinkingProcessPayload {
+    /**
+     * Markdown chain-of-thought text.
+     */
+    body: string;
+    defaultExpanded?: boolean;
+    [k: string]: any;
+}

package/dist/generated/ThinkingProcessPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/generated/ToolExecutionPayload.d.ts

@@ -0,0 +1,28 @@
+/**
+ * Payload of a tool-execution card. Unified representation of a single tool / MCP / local action invocation: name, status, args (input), output, and an optional human-readable summary. Both deeppath's ActionSegment and deeppath-agent's ExecutedActionsCard render off this same shape.
+ */
+export interface ToolExecutionPayload {
+    /**
+     * Stable id of this invocation.
+     */
+    id: string;
+    /**
+     * Tool name as exposed to the agent.
+     */
+    name: string;
+    status: "pending" | "running" | "succeeded" | "failed" | "cancelled";
+    /**
+     * One-line human summary for the row header.
+     */
+    summary?: string | null;
+    args?: any;
+    output?: any;
+    error?: string | null;
+    durationMs?: number | null;
+    /**
+     * Optional icon hint (lucide name or url).
+     */
+    icon?: string | null;
+    expandable?: boolean;
+    [k: string]: any;
+}

package/dist/generated/ToolExecutionPayload.js

@@ -0,0 +1 @@
+export {};

package/dist/safety-patterns.d.ts

@@ -1,3 +1,26 @@
+/**
+ * Shell command safety patterns — classify a command before it runs.
+ *
+ * Full 61-rule set reflowed from deeppath-agent's
+ * `src/harness/safety-patterns.ts`; the framework is now the source of truth
+ * (agent will import from here in A4). Kept in lockstep with the Python twin
+ * `packages/agent-harness/py/src/steerable_agent_harness/safety.py` via the
+ * conformance case `tests/conformance/cases/safety/classify_shell_command.yaml`.
+ *
+ * Rule order matters: `matchedRules` follows BUILTIN_PATTERNS order so both
+ * languages return identical lists.
+ */
+export interface CommandSafetyConfig {
+    disabledPatternIds?: string[];
+    hiddenPatternIds?: string[];
+    customPatterns?: Array<{
+        id: string;
+        label: string;
+        pattern: string;
+        category: string;
+        enabled: boolean;
+    }>;
+}
 export type PatternSeverity = "critical" | "warning";
 export type PatternPlatform = "all" | "unix" | "windows";
 export interface SafetyPatternDef {
@@ -9,8 +32,22 @@ export interface SafetyPatternDef {
     severity: PatternSeverity;
     platform: PatternPlatform;
 }
+export declare const SAFETY_CATEGORIES: Record<string, string>;
 export declare const BUILTIN_PATTERNS: SafetyPatternDef[];
-export declare function classifyShellCommand(command: string): {
+/**
+ * 根据用户配置计算当前生效的正则模式列表
+ */
+export declare function getActivePatterns(config?: CommandSafetyConfig | null): RegExp[];
+/**
+ * 按分类分组返回内置模式
+ */
+export declare function getPatternsByCategory(): Record<string, SafetyPatternDef[]>;
+/**
+ * 默认空配置（所有内置模式均启用，无自定义模式）
+ */
+export declare const DEFAULT_COMMAND_SAFETY_CONFIG: CommandSafetyConfig;
+export interface ShellCommandClassification {
     severity: "safe" | PatternSeverity;
     matchedRules: string[];
-};
+}
+export declare function classifyShellCommand(command: string, config?: CommandSafetyConfig | null): ShellCommandClassification;

package/dist/safety-patterns.js

@@ -1,71 +1,166 @@
+/**
+ * Shell command safety patterns — classify a command before it runs.
+ *
+ * Full 61-rule set reflowed from deeppath-agent's
+ * `src/harness/safety-patterns.ts`; the framework is now the source of truth
+ * (agent will import from here in A4). Kept in lockstep with the Python twin
+ * `packages/agent-harness/py/src/steerable_agent_harness/safety.py` via the
+ * conformance case `tests/conformance/cases/safety/classify_shell_command.yaml`.
+ *
+ * Rule order matters: `matchedRules` follows BUILTIN_PATTERNS order so both
+ * languages return identical lists.
+ */
+// ─── Categories ──────────────────────────────────────────────────────
+export const SAFETY_CATEGORIES = {
+    file_ops: "文件操作",
+    system: "系统管理",
+    process: "进程管理",
+    network: "网络安全",
+    package: "包管理",
+    vcs: "版本控制",
+    container: "容器管理",
+    file_write: "文件写入",
+    windows: "Windows / PowerShell",
+};
+// ─── Built-in patterns ───────────────────────────────────────────────
+// Merged from deeppath-agent local-actions.ts (web, severity=warning) and
+// local-executor.ts (desktop, severity=critical). Keep aligned with the
+// Python twin.
 export const BUILTIN_PATTERNS = [
-    {
-        id: "rm",
-        label: "rm 删除",
-        description: "匹配 rm 命令",
-        pattern: "\\brm\\s",
-        category: "file_ops",
-        severity: "warning",
-        platform: "unix",
-    },
-    {
-        id: "rm_rf_root",
-        label: "rm -rf /",
-        description: "递归删除根目录",
-        pattern: "rm\\s+-rf\\s+\\/(?:\\s|$)",
-        category: "file_ops",
-        severity: "critical",
-        platform: "unix",
-    },
-    {
-        id: "sudo",
-        label: "sudo 提权",
-        description: "以超级用户权限执行",
-        pattern: "\\bsudo\\s",
-        category: "system",
-        severity: "critical",
-        platform: "unix",
-    },
-    {
-        id: "curl_pipe_sh",
-        label: "curl | sh",
-        description: "从网络下载并直接执行脚本",
-        pattern: "\\bcurl\\s.*\\|\\s*(sh|bash|zsh)",
-        category: "network",
-        severity: "warning",
-        platform: "unix",
-    },
-    {
-        id: "git_push",
-        label: "git push / reset --hard",
-        description: "Git 远程推送或硬重置",
-        pattern: "\\bgit\\s+(push|reset\\s+--hard|clean\\s+-fd)",
-        category: "vcs",
-        severity: "warning",
-        platform: "all",
-    },
-    {
-        id: "win_del_force",
-        label: "del /f /s /q 强制删除",
-        description: "强制递归删除整个驱动器",
-        pattern: "\\bdel\\s+\\/f\\s+\\/s\\s+\\/q\\s+[a-z]:\\\\",
-        category: "windows",
-        severity: "critical",
-        platform: "windows",
-    },
+    // ── file_ops ──
+    { id: "rm", label: "rm 删除", description: "匹配 rm 命令", pattern: "\\brm\\s", category: "file_ops", severity: "warning", platform: "unix" },
+    { id: "rm_end", label: "rm（行尾）", description: "匹配行尾的 rm", pattern: "\\brm$", category: "file_ops", severity: "warning", platform: "unix" },
+    { id: "rm_rf_root", label: "rm -rf /", description: "递归删除根目录", pattern: "rm\\s+-rf\\s+\\/(?:\\s|$)", category: "file_ops", severity: "critical", platform: "unix" },
+    { id: "rmdir", label: "rmdir 删除目录", description: "匹配 rmdir 命令", pattern: "\\brmdir\\s", category: "file_ops", severity: "warning", platform: "unix" },
+    { id: "mv", label: "mv 移动/重命名", description: "匹配 mv 命令", pattern: "\\bmv\\s.*\\/", category: "file_ops", severity: "warning", platform: "unix" },
+    { id: "shred", label: "shred 安全擦除", description: "不可恢复地擦除文件", pattern: "\\bshred\\b", category: "file_ops", severity: "warning", platform: "unix" },
+    { id: "truncate", label: "truncate 截断文件", description: "截断文件内容", pattern: "\\btruncate\\b", category: "file_ops", severity: "warning", platform: "unix" },
+    // ── system ──
+    { id: "sudo", label: "sudo 提权", description: "以超级用户权限执行", pattern: "\\bsudo\\s", category: "system", severity: "critical", platform: "unix" },
+    { id: "mkfs", label: "mkfs 格式化磁盘", description: "创建文件系统（格式化）", pattern: "\\bmkfs\\b", category: "system", severity: "critical", platform: "unix" },
+    { id: "dd", label: "dd 磁盘写入", description: "底层磁盘数据复制", pattern: "\\bdd\\s", category: "system", severity: "warning", platform: "unix" },
+    { id: "dd_if", label: "dd if= 磁盘镜像", description: "使用 dd if= 读写磁盘", pattern: "\\bdd\\s+if=", category: "system", severity: "critical", platform: "unix" },
+    { id: "format", label: "format 格式化", description: "格式化磁盘", pattern: "\\bformat\\b", category: "system", severity: "warning", platform: "all" },
+    { id: "shutdown", label: "shutdown 关机", description: "关闭系统", pattern: "\\bshutdown\\b", category: "system", severity: "warning", platform: "all" },
+    { id: "reboot", label: "reboot 重启", description: "重启系统", pattern: "\\breboot\\b", category: "system", severity: "warning", platform: "all" },
+    { id: "chmod", label: "chmod 修改权限", description: "修改文件权限", pattern: "\\bchmod\\s", category: "system", severity: "warning", platform: "unix" },
+    { id: "chmod_777_root", label: "chmod -R 777 /", description: "递归赋予根目录所有权限", pattern: "chmod\\s+-R\\s+777\\s+\\/(?:\\s|$)", category: "system", severity: "critical", platform: "unix" },
+    { id: "chown", label: "chown 修改所有者", description: "修改文件所有者", pattern: "\\bchown\\s", category: "system", severity: "warning", platform: "unix" },
+    { id: "fork_bomb", label: "Fork Bomb", description: ":(){ :|:& };: fork 炸弹", pattern: ":\\(\\)\\s*\\{\\s*:\\|:&\\s*\\};:", category: "system", severity: "critical", platform: "unix" },
+    // ── process ──
+    { id: "kill", label: "kill 终止进程", description: "向进程发送信号", pattern: "\\bkill\\s", category: "process", severity: "warning", platform: "unix" },
+    { id: "killall", label: "killall 终止所有", description: "按名称终止进程", pattern: "\\bkillall\\s", category: "process", severity: "warning", platform: "unix" },
+    // ── network ──
+    { id: "curl_pipe_sh", label: "curl | sh", description: "从网络下载并直接执行脚本", pattern: "\\bcurl\\s.*\\|\\s*(sh|bash|zsh)", category: "network", severity: "warning", platform: "unix" },
+    { id: "wget_pipe_sh", label: "wget | sh", description: "从网络下载并直接执行脚本", pattern: "\\bwget\\s.*\\|\\s*(sh|bash|zsh)", category: "network", severity: "warning", platform: "unix" },
+    { id: "redirect_dev", label: "重定向到 /dev/", description: "向设备文件写入数据", pattern: ">\\s*\\/dev\\/", category: "network", severity: "warning", platform: "unix" },
+    // ── package ──
+    { id: "npm_publish", label: "npm publish", description: "发布/取消发布 npm 包", pattern: "\\bnpm\\s+(publish|unpublish)", category: "package", severity: "warning", platform: "all" },
+    { id: "pip_install", label: "pip install", description: "安装 Python 包", pattern: "\\bpip\\s+install\\b", category: "package", severity: "warning", platform: "all" },
+    { id: "npm_install", label: "npm install", description: "安装 npm 包", pattern: "\\bnpm\\s+install\\b", category: "package", severity: "warning", platform: "all" },
+    { id: "yarn_add", label: "yarn add", description: "添加 yarn 依赖", pattern: "\\byarn\\s+add\\b", category: "package", severity: "warning", platform: "all" },
+    { id: "pnpm_add", label: "pnpm add", description: "添加 pnpm 依赖", pattern: "\\bpnpm\\s+add\\b", category: "package", severity: "warning", platform: "all" },
+    { id: "uv_add", label: "uv add", description: "添加 uv 依赖", pattern: "\\buv\\s+add\\b", category: "package", severity: "warning", platform: "all" },
+    { id: "apt_install", label: "apt install/remove", description: "系统包管理器操作", pattern: "\\bapt(-get)?\\s+(install|remove|purge)", category: "package", severity: "warning", platform: "unix" },
+    { id: "brew_install", label: "brew install/uninstall", description: "Homebrew 包管理器操作", pattern: "\\bbrew\\s+(install|uninstall|remove)", category: "package", severity: "warning", platform: "unix" },
+    // ── vcs ──
+    { id: "git_push", label: "git push / reset --hard", description: "Git 远程推送或硬重置", pattern: "\\bgit\\s+(push|reset\\s+--hard|clean\\s+-fd)", category: "vcs", severity: "warning", platform: "all" },
+    // ── container ──
+    { id: "docker_rm", label: "docker rm/rmi/prune", description: "删除容器/镜像或清理系统", pattern: "\\bdocker\\s+(rm|rmi|system\\s+prune)", category: "container", severity: "warning", platform: "all" },
+    // ── file_write ──
+    { id: "redirect_overwrite", label: "> / >> 重定向写入", description: "文件重定向覆盖或追加", pattern: "\\b(>\\s|>>)\\s*[^|]", category: "file_write", severity: "warning", platform: "all" },
+    { id: "tee", label: "tee 写入文件", description: "将输出写入文件", pattern: "\\btee\\s", category: "file_write", severity: "warning", platform: "unix" },
+    { id: "sed_inplace", label: "sed -i 原地修改", description: "直接修改文件内容", pattern: "\\bsed\\s+-i", category: "file_write", severity: "warning", platform: "unix" },
+    // ── windows ──
+    { id: "win_del", label: "del 删除", description: "Windows 删除命令", pattern: "\\bdel\\s", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_rd", label: "rd 删除目录", description: "Windows 删除目录", pattern: "\\brd\\s", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_rd_end", label: "rd（行尾）", description: "匹配行尾的 rd", pattern: "\\brd$", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_rdel", label: "rdel", description: "Windows rdel 命令", pattern: "\\brdel\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_del_force", label: "del /f /s /q 强制删除", description: "强制递归删除整个驱动器", pattern: "\\bdel\\s+\\/f\\s+\\/s\\s+\\/q\\s+[a-z]:\\\\", category: "windows", severity: "critical", platform: "windows" },
+    { id: "win_rd_force", label: "rd /s /q 强制删除目录", description: "强制递归删除整个驱动器目录", pattern: "\\brd\\s+\\/s\\s+\\/q\\s+[a-z]:\\\\", category: "windows", severity: "critical", platform: "windows" },
+    { id: "win_remove_item", label: "Remove-Item", description: "PowerShell 删除项", pattern: "\\bRemove-Item\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_stop_process", label: "Stop-Process", description: "PowerShell 终止进程", pattern: "\\bStop-Process\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_stop_computer", label: "Stop-Computer", description: "PowerShell 关机", pattern: "\\bStop-Computer\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_restart_computer", label: "Restart-Computer", description: "PowerShell 重启", pattern: "\\bRestart-Computer\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_set_execution_policy", label: "Set-ExecutionPolicy", description: "修改脚本执行策略", pattern: "\\bSet-ExecutionPolicy\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_format_volume", label: "Format-Volume", description: "PowerShell 格式化卷", pattern: "\\bFormat-Volume\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_clear_disk", label: "Clear-Disk", description: "PowerShell 清除磁盘", pattern: "\\bClear-Disk\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_wsl", label: "wsl 子系统", description: "调用 WSL 子系统", pattern: "\\bwsl\\s", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_powershell_cmd", label: "powershell -Command", description: "通过 PowerShell 执行命令", pattern: "\\bpowershell\\s.*-[Cc]ommand", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_pwsh", label: "pwsh", description: "PowerShell Core", pattern: "\\bpwsh\\s", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_cmd_c", label: "cmd /c", description: "CMD 执行命令", pattern: "\\bcmd\\s*\\/c\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_reg", label: "reg delete/add", description: "注册表操作", pattern: "\\breg\\s+(delete|add)\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_net", label: "net user/stop/start", description: "网络和用户管理", pattern: "\\bnet\\s+(user|stop|start)\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_sc", label: "sc delete/stop/config", description: "服务管理", pattern: "\\bsc\\s+(delete|stop|config)\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_diskpart", label: "diskpart", description: "磁盘分区工具", pattern: "\\bdiskpart\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_bcdedit", label: "bcdedit", description: "启动配置编辑", pattern: "\\bbcdedit\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_sfc", label: "sfc", description: "系统文件检查器", pattern: "\\bsfc\\b", category: "windows", severity: "warning", platform: "windows" },
+    { id: "win_dism", label: "dism", description: "部署映像服务和管理", pattern: "\\bdism\\b", category: "windows", severity: "warning", platform: "windows" },
+    // 只匹配 "format <盘符>:" 形式的磁盘格式化（含 format.com），避免误伤
+    // PowerShell 的 Format-List / Format-Table 等格式化输出 cmdlet
+    // （Format-Volume 由上面的 win_format_volume 单独覆盖）。
+    { id: "win_format_cmd", label: "format（Windows）", description: "Windows 格式化磁盘命令", pattern: "\\bformat(\\.com)?\\s+[a-z]:", category: "windows", severity: "critical", platform: "windows" },
 ];
-export function classifyShellCommand(command) {
+// ─── Helpers ─────────────────────────────────────────────────────────
+/**
+ * 根据用户配置计算当前生效的正则模式列表
+ */
+export function getActivePatterns(config) {
+    const disabled = new Set(config?.disabledPatternIds ?? []);
+    const patterns = BUILTIN_PATTERNS.filter((p) => !disabled.has(p.id)).map((p) => {
+        const flags = p.platform === "windows" ? "i" : undefined;
+        return new RegExp(p.pattern, flags);
+    });
+    if (config?.customPatterns) {
+        for (const cp of config.customPatterns) {
+            if (!cp.enabled)
+                continue;
+            try {
+                patterns.push(new RegExp(cp.pattern));
+            }
+            catch {
+                // skip invalid regex
+            }
+        }
+    }
+    return patterns;
+}
+/**
+ * 按分类分组返回内置模式
+ */
+export function getPatternsByCategory() {
+    const grouped = {};
+    for (const p of BUILTIN_PATTERNS) {
+        if (!grouped[p.category])
+            grouped[p.category] = [];
+        grouped[p.category].push(p);
+    }
+    return grouped;
+}
+/**
+ * 默认空配置（所有内置模式均启用，无自定义模式）
+ */
+export const DEFAULT_COMMAND_SAFETY_CONFIG = {
+    disabledPatternIds: [],
+    hiddenPatternIds: [],
+    customPatterns: [],
+};
+export function classifyShellCommand(command, config) {
     const normalized = command.trim();
-    if (!normalized)
+    if (!normalized) {
         return { severity: "safe", matchedRules: [] };
-    const matches = BUILTIN_PATTERNS.filter((rule) => {
+    }
+    const disabled = new Set(config?.disabledPatternIds ?? []);
+    const matches = BUILTIN_PATTERNS.filter((rule) => !disabled.has(rule.id)).filter((rule) => {
         const flags = rule.platform === "windows" ? "i" : undefined;
         return new RegExp(rule.pattern, flags).test(normalized);
     });
-    if (!matches.length)
+    if (!matches.length) {
         return { severity: "safe", matchedRules: [] };
+    }
+    const severity = matches.some((item) => item.severity === "critical") ? "critical" : "warning";
     return {
-        severity: matches.some((m) => m.severity === "critical") ? "critical" : "warning",
-        matchedRules: matches.map((m) => m.id),
+        severity,
+        matchedRules: matches.map((item) => item.id),
     };
 }

package/package.json

@@ -1,9 +1,9 @@
 {
   "name": "@steerable/agent-harness",
-  "version": "0.2.3",
+  "version": "0.2.5",
   "description": "Steerable framework Tier 2 — TypeScript facade over the Python harness (policy, budget, retry, completion, tracing) used for cross-language conformance tests. Production code should depend on the Python harness.",
   "license": "Apache-2.0",
-  "homepage": "https://pathlyapp.github.io/steerable-framework/",
+  "homepage": "https://steerableframework.com/",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/pathlyapp/steerable-framework.git",
@@ -29,7 +29,7 @@
     "README.md"
   ],
   "dependencies": {
-    "@steerable/agent-protocol": "0.2.3"
+    "@steerable/agent-protocol": "0.2.5"
   },
   "devDependencies": {
     "typescript": "^5.8.3",