@botconnector/bccli 0.4.0 → 0.4.2

package/README.en.md

@@ -19,6 +19,9 @@ edits files, and runs commands — with your permission.
 # one-line install (macOS / Linux / WSL2) — provisions a portable Node if needed, no sudo
 curl -fsSL https://raw.githubusercontent.com/farijarifriyanto-debug/bccli/main/install.sh | bash
 
+# Windows (PowerShell 5.1+) — needs Node.js 22+ already installed
+irm https://raw.githubusercontent.com/farijarifriyanto-debug/bccli/main/install.ps1 | iex
+
 # or manual, if you already have Node.js 22+
 npm i -g @botconnector/bccli        # latest stable, requires Node.js 22+
 npm i -g @botconnector/bccli@next   # …or the next prerelease
@@ -53,13 +56,18 @@ In-session commands (type `/` for suggestions, ↑↓ to select, Enter to run):
 | `/session` | info about this session |
 | `/status` | version, model, permission mode, MCP, context |
 | `/permissions` | view/revoke permissions |
+| `/pr` | review a pull request (`/pr <n\|url\|branch>`) |
+| `/worktree` | list git worktrees; create one with `bccli -w <name>` |
 | `/undo` | undo the last file-edit turn |
+| `/redo` | re-apply the file edits that /undo just reverted |
+| `/rewind` | step back turns: conversation + file edits (`/rewind 2`) |
 | `/diff` | git diff of the project |
 | `/copy` | copy the last answer |
 | `/export` | save the conversation to markdown |
 | `/memory` | project instructions (/memory <text>, /memory global <text>) |
 | `/init` | create/update AGENTS.md for this project |
 | `/agents` | list subagents |
+| `/tasks` | list background tasks |
 | `/skills` | list skills and custom commands |
 | `/doctor` | check installation health |
 | `/login` | store the active provider's API key |
@@ -125,6 +133,8 @@ the `reasoning` preference is read only from the user config `~/.bccli/config.js
 cloned repo — permissions (`permissionMode`, `allow`) and existing providers are configured only
 from `~/.bccli/config.json`, so a cloned repo cannot grant itself full permissions or steal API keys.
 
+Every key of `~/.bccli/config.json` with its values and scope: [`docs/config.md`](docs/config.md).
+
 ## External agent integrations
 
 BCCLI can set up BotConnector as a provider for other agents:
@@ -221,16 +231,17 @@ Tasks with ≥ 3 steps show a checklist above the input box. In plan mode (Shift
 the agent only reads and then proposes a plan; `[a]` approves with automatic edits,
 `[y]` approves asking at each step, `[n]` asks for revisions.
 
-Project instructions are read from `AGENTS.md` and `BCCLI.md`.
+Project instructions are read from `AGENTS.md`, `CLAUDE.md`, and `BCCLI.md`.
 
 ## Development
 
 ```bash
 npm install
 npm test && npm run typecheck && npm run lint && npm run build
+npm run coverage   # coverage gate (also enforced on Linux CI)
 ```
 
-CI (`ci.yml`) runs the same on Linux, macOS, and Windows.
+CI (`ci.yml`) runs the same on Linux, macOS, and Windows, plus the coverage gate on Linux.
 Releases: bump `version` in `package.json`, then push a `v<version>` tag — `release.yml`
 runs tests, a smoke test against BotConnector Cloud, and `npm publish --provenance`.
 

package/README.md

@@ -19,6 +19,9 @@ dan menjalankan perintah — dengan izin kamu.
 # install satu baris (macOS / Linux / WSL2) — pasang Node portable bila perlu, tanpa sudo
 curl -fsSL https://raw.githubusercontent.com/farijarifriyanto-debug/bccli/main/install.sh | bash
 
+# Windows (PowerShell 5.1+) — butuh Node.js 22+ terpasang lebih dulu
+irm https://raw.githubusercontent.com/farijarifriyanto-debug/bccli/main/install.ps1 | iex
+
 # atau manual, bila Node.js 22+ sudah terpasang
 npm i -g @botconnector/bccli        # rilis stabil terbaru, butuh Node.js 22+
 npm i -g @botconnector/bccli@next   # …atau pra-rilis berikutnya
@@ -53,13 +56,18 @@ Perintah di dalam sesi (ketik `/` untuk saran, ↑↓ pilih, Enter jalankan):
 | `/session` | info sesi ini |
 | `/status` | versi, model, mode izin, MCP, konteks |
 | `/permissions` | lihat/cabut izin |
+| `/pr` | review pull request (`/pr <n\|url\|branch>`) |
+| `/worktree` | daftar git worktree; buat dengan `bccli -w <nama>` |
 | `/undo` | batalkan edit file giliran terakhir |
+| `/redo` | terapkan ulang edit file yang baru dibatalkan `/undo` |
+| `/rewind` | mundur beberapa giliran: percakapan + edit file (`/rewind 2`) |
 | `/diff` | git diff project |
 | `/copy` | salin jawaban terakhir |
 | `/export` | simpan percakapan ke markdown |
 | `/memory` | instruksi project (/memory <teks>, /memory global <teks>) |
 | `/init` | buat/perbarui AGENTS.md untuk project ini |
 | `/agents` | daftar subagent |
+| `/tasks` | daftar tugas latar belakang |
 | `/skills` | daftar skill dan perintah custom |
 | `/doctor` | cek kesehatan instalasi |
 | `/login` | simpan API key provider aktif |
@@ -124,6 +132,8 @@ preference `reasoning` hanya dibaca dari config user `~/.bccli/config.json`, buk
 izin (`permissionMode`, `allow`) dan provider yang sudah ada hanya diatur dari `~/.bccli/config.json`,
 supaya repo yang kamu clone tidak bisa memberi dirinya izin penuh atau mencuri API key.
 
+Daftar lengkap kunci `~/.bccli/config.json` beserta nilai dan cakupannya: [`docs/config.md`](docs/config.md).
+
 ## Integrasi agent eksternal
 
 BCCLI dapat memasang BotConnector sebagai provider untuk agent lain:
@@ -218,16 +228,17 @@ Tugas ≥ 3 langkah menampilkan checklist di atas kotak input. Di mode plan (Shi
 agent hanya membaca lalu mengajukan rencana; `[a]` setujui dengan edit otomatis,
 `[y]` setujui dengan tanya tiap langkah, `[n]` minta perbaikan.
 
-Instruksi project dibaca dari `AGENTS.md` dan `BCCLI.md`.
+Instruksi project dibaca dari `AGENTS.md`, `CLAUDE.md`, dan `BCCLI.md`.
 
 ## Pengembangan
 
 ```bash
 npm install
 npm test && npm run typecheck && npm run lint && npm run build
+npm run coverage   # gerbang coverage (juga dijalankan di CI Linux)
 ```
 
-CI (`ci.yml`) menjalankan hal yang sama di Linux, macOS, dan Windows.
+CI (`ci.yml`) menjalankan hal yang sama di Linux, macOS, dan Windows, plus gerbang coverage di Linux.
 Rilis: naikkan `version` di `package.json`, lalu push tag `v<version>` — `release.yml`
 menjalankan test, smoke test ke BotConnector Cloud, dan `npm publish --provenance`.
 

package/dist/abort.d.ts

@@ -0,0 +1,5 @@
+export declare const ABORTED: unique symbol;
+/** Resolves with the promise's value, or with ABORTED as soon as the signal fires. */
+export declare function raceAbort<T>(promise: Promise<T>, signal: AbortSignal): Promise<T | typeof ABORTED>;
+/** Collects an async iterable, giving up (and telling the iterator to stop) as soon as the signal fires. */
+export declare function collectUntilAbort<T>(iterable: AsyncIterable<T>, signal: AbortSignal | undefined, onItem: (item: T) => void): Promise<boolean>;

package/dist/{acp-ATIL4JOZ.js → acp-UYC6APR5.js}

@@ -36,6 +36,9 @@ function runAcp(input, output, opts) {
   const replyError = (id, code, message) => write({ jsonrpc: "2.0", id, error: { code, message } });
   const update = (sessionId, body) => write({ jsonrpc: "2.0", method: "session/update", params: { sessionId, update: body } });
   const ioFor = (sid, session) => ({
+    onThinking(delta) {
+      update(sid, { sessionUpdate: "agent_thought_chunk", content: { type: "text", text: delta } });
+    },
     toolStart(id, tool, target) {
       session.open.push({ id, tool, target });
       update(sid, { sessionUpdate: "tool_call", toolCallId: id, title: titleOf(tool, target), kind: kindOf(tool), status: "in_progress", rawInput: { target } });
@@ -89,7 +92,7 @@ function runAcp(input, output, opts) {
         protocolVersion: ACP_PROTOCOL_VERSION,
         agentCapabilities: {
           loadSession: false,
-          promptCapabilities: { image: false, audio: false, embeddedContext: false }
+          promptCapabilities: { image: false, audio: false, embeddedContext: true }
         }
       });
     }
@@ -108,7 +111,17 @@ function runAcp(input, output, opts) {
       const sid = params?.sessionId ?? "";
       const session = sessions.get(sid);
       if (!session) return replyError(id, -32602, "Unknown session");
-      const text = (params?.prompt ?? []).filter((p) => p?.type === "text").map((p) => String(p.text ?? "")).join("\n");
+      if (session.controller) return replyError(id, -32e3, "Session is busy");
+      const text = (params?.prompt ?? []).map((p) => {
+        if (p?.type === "text") return String(p.text ?? "");
+        if ((p?.type === "resource" || p?.type === "embeddedContext") && p.resource?.text) {
+          const label = p.resource.uri ? `
+[${p.resource.uri}]
+` : "\n";
+          return `${label}${p.resource.text}`;
+        }
+        return "";
+      }).filter(Boolean).join("\n");
       const controller = new AbortController();
       session.controller = controller;
       let cancelled = false;
@@ -122,8 +135,10 @@ function runAcp(input, output, opts) {
         if (cancelled) return reply(id, { stopReason: "cancelled" });
         return replyError(id, -32603, e.message);
       } finally {
-        session.controller = null;
-        session.open = [];
+        if (session.controller === controller) {
+          session.controller = null;
+          session.open = [];
+        }
       }
       return reply(id, { stopReason: cancelled ? "cancelled" : "end_turn" });
     }
@@ -150,6 +165,10 @@ function runAcp(input, output, opts) {
     };
     const onEnd = () => {
       input.removeListener("data", onData);
+      for (const session of sessions.values()) {
+        session.controller?.abort();
+        for (const settle of [...session.pending]) settle("no");
+      }
       resolve();
     };
     input.on("data", onData);
@@ -161,17 +180,22 @@ function runtimeAgent(cwd, allowAll = false) {
   return {
     async run(text, onChunk, signal, io) {
       if (!rt) {
-        const { parseCliArgs } = await import("./args-J5ONVGIF.js");
-        const { createRuntime } = await import("./setup-FJAWDWV4.js");
+        const { parseCliArgs } = await import("./args-NRTZWWEV.js");
+        const { createRuntime } = await import("./setup-S4AJT6XE.js");
         rt = createRuntime({ cwd, args: { ...parseCliArgs([]), allowAll } });
       }
       if (io && !allowAll) rt.agent.askPermission = (req) => io.askPermission(req);
+      let turnError;
       rt.agent.onEvent = (event) => {
         if (event.type === "text") onChunk(event.delta);
+        else if (event.type === "thinking") io?.onThinking?.(event.delta);
         else if (event.type === "toolStart") io?.toolStart(event.id, event.tool, event.target);
         else if (event.type === "toolEnd") io?.toolEnd(event.id, event.output, event.isError);
+        else if (event.type === "error") turnError = event.message;
+        else if (event.type === "budgetExceeded") turnError = `Budget exceeded (${event.kind})`;
       };
       await rt.agent.run(text, signal);
+      if (turnError) throw new Error(turnError);
     }
   };
 }

package/dist/acp.d.ts

@@ -0,0 +1,28 @@
+import type { PermissionAnswer, PermissionAsk } from './agent';
+export declare const ACP_PROTOCOL_VERSION = 1;
+/** What the agent reports to the editor besides text: tool calls, and permission questions the editor answers. */
+export interface AcpIo {
+    toolStart(id: string, tool: string, target: string): void;
+    toolEnd(id: string, output: string, isError: boolean): void;
+    askPermission(req: PermissionAsk): Promise<PermissionAnswer>;
+    onThinking?(delta: string): void;
+}
+/** The minimum an ACP client needs: run one prompt, stream text chunks, allow cancellation. */
+export interface AcpAgent {
+    run(text: string, onChunk: (text: string) => void, signal: AbortSignal, io?: AcpIo): Promise<void>;
+}
+/**
+ * Agent Client Protocol over newline-delimited JSON-RPC (the transport editors like Zed use).
+ * Implements the subset: initialize, session/new, session/prompt (streaming session/update
+ * notifications: text chunks and tool calls), session/request_permission, session/cancel;
+ * anything else with an id gets -32601.
+ */
+export declare function runAcp(input: NodeJS.ReadableStream, output: NodeJS.WritableStream, opts: {
+    cwd: string;
+    newAgent: (cwd: string) => AcpAgent;
+}): Promise<void>;
+/**
+ * Production agent: one bccli runtime per ACP session. Permission questions go to the editor
+ * (session/request_permission); `allowAll` skips them and runs everything unasked.
+ */
+export declare function runtimeAgent(cwd: string, allowAll?: boolean): AcpAgent;

package/dist/agent.d.ts

@@ -0,0 +1,147 @@
+import type { PermissionRequest, Permissions } from './permissions';
+import { type ChatMessage, type Provider, type Usage } from './provider';
+import type { ReasoningLevel } from './reasoning';
+import { type HookEvent, type HooksConfig } from './hooks';
+import type { Tool } from './tools/types';
+import { type UsageCap } from './budget';
+export type AgentEvent = {
+    type: 'text';
+    delta: string;
+} | {
+    type: 'thinking';
+    delta: string;
+} | {
+    type: 'textReplace';
+    text: string;
+} | {
+    type: 'toolStart';
+    id: string;
+    tool: string;
+    target: string;
+} | {
+    type: 'toolEnd';
+    id: string;
+    tool: string;
+    output: string;
+    display?: string;
+    isError: boolean;
+} | {
+    type: 'usage';
+    inputTokens: number;
+    outputTokens: number;
+    cachedInputTokens?: number;
+    cacheWriteTokens?: number;
+} | {
+    type: 'compacted';
+} | {
+    type: 'stepLimit';
+    maxSteps: number;
+} | {
+    type: 'budgetExceeded';
+    kind: 'tokens' | 'usd';
+    used: number;
+    limit: number;
+} | {
+    type: 'aborted';
+} | {
+    type: 'error';
+    message: string;
+} | {
+    type: 'done';
+} | {
+    type: 'subagent';
+    parentId: string;
+    agent: string;
+    event: AgentEvent;
+};
+export type PermissionAnswer = 'yes' | 'session' | 'all' | 'no';
+export type PermissionAsk = PermissionRequest & {
+    preview?: string;
+    sessionRules?: string[];
+    agent?: string;
+};
+export type AskPermission = (req: PermissionAsk) => Promise<PermissionAnswer>;
+export interface AgentOptions {
+    provider: Provider;
+    tools: Tool[];
+    permissions: Permissions;
+    systemPrompt: string;
+    cwd: string;
+    history?: ChatMessage[];
+    maxSteps?: number | null;
+    contextWindow?: number;
+    onMessage?: (message: ChatMessage) => void;
+    onReset?: () => void;
+    /** Per-turn token usage delta (provider-reported or estimated); the session persists it for stats. */
+    onUsage?: (usage: Usage) => void;
+    /** Marks the start of a user turn (undo boundary). */
+    onTurnStart?: () => void;
+    checkpoint?: (absPath: string) => Promise<void>;
+    /** Subagent name, shown on its permission prompts. */
+    label?: string;
+    reasoning?: ReasoningLevel;
+    /** Tool/session hooks (global config only). */
+    hooks?: HooksConfig;
+    /** In-process plugin observers fired alongside the shell hooks (global config plugins). */
+    pluginEmit?: (event: HookEvent, payload?: {
+        tool?: string;
+        input?: unknown;
+        output?: string;
+    }) => Promise<void>;
+    /** Base environment handed to hook processes (defaults to the process env at spawn). */
+    env?: NodeJS.ProcessEnv;
+    /** Per-session token/USD budget (global config usageCap); blocks model calls once exceeded. */
+    usageCap?: UsageCap;
+    /** Model ref used to match usageCap.prices keys. */
+    budgetModel?: string;
+}
+export declare class Agent {
+    private readonly opts;
+    provider: Provider;
+    permissions: Permissions;
+    messages: ChatMessage[];
+    totalUsage: Usage;
+    lastInputTokens: number;
+    onEvent: (event: AgentEvent) => void;
+    reasoning: ReasoningLevel;
+    askPermission: AskPermission;
+    tools: Tool[];
+    private definitions;
+    private turnTools;
+    private turnDefinitions;
+    readonly readFiles: Set<string>;
+    private readonly webBudget;
+    private readonly maxSteps?;
+    private contextWindow;
+    private budgetModel;
+    private systemPrompt;
+    constructor(opts: AgentOptions);
+    setSystemPrompt(text: string): void;
+    setContextWindow(window: number): void;
+    getContextWindow(): number;
+    setBudgetModel(model: string): void;
+    getBudgetModel(): string;
+    setTools(tools: Tool[]): void;
+    private isParallelSafe;
+    private push;
+    /** Swaps in another conversation (resume, new session) without writing to the session log. */
+    load(messages: ChatMessage[]): void;
+    clear(): void;
+    compact(signal: AbortSignal): Promise<void>;
+    /**
+     * Events of one tool execution. A cancelled turn gives up on a tool that does not stop, and whatever that tool
+     * reports later must not reach the screen or a newer turn, so this is per execution and not per call id (ids repeat).
+     */
+    private report;
+    run(text: string, signal: AbortSignal): Promise<void>;
+    /** Current cap status; undefined means the budget is not exhausted. */
+    private overBudget;
+    /** Every tool call needs an answer in the history, or the next request to the model is rejected. */
+    private cancelCalls;
+    /**
+     * Every tool call must resolve with text: a rejected one would leave the assistant tool_calls
+     * without answers, and the next request to the model would be rejected until a restart.
+     */
+    private runTool;
+    private executeTool;
+}

package/dist/agents.d.ts

@@ -0,0 +1,4 @@
+import type { AgentDef } from './extensions';
+export declare const EXPLORE_PROMPT = "You are a read-only research agent working for another agent. Search and read the codebase to answer the request. You cannot edit files or run commands. Reply with a concise report: the answer, the relevant file paths with line numbers, and nothing else.";
+export declare const GENERAL_PROMPT = "You are an agent handling a delegated task for another agent. Complete the task end to end with the tools you have, verify the result, and reply with a short report of what you did and what you found.";
+export declare const BUILTIN_AGENTS: AgentDef[];

package/dist/args-NRTZWWEV.js

@@ -0,0 +1,13 @@
+#!/usr/bin/env node
+import {
+  helpText,
+  parseCliArgs
+} from "./chunk-6UO5GX55.js";
+import "./chunk-WFZKJY3R.js";
+import "./chunk-KPTML72Q.js";
+import "./chunk-RX4UQX43.js";
+import "./chunk-3WHI4LN3.js";
+export {
+  helpText,
+  parseCliArgs
+};

package/dist/args.d.ts

@@ -0,0 +1,53 @@
+import { type PermissionMode } from './config';
+import { type ReasoningLevel } from './reasoning';
+import { type Lang } from './i18n';
+export interface CliArgs {
+    command: 'run' | 'login' | 'models' | 'update' | 'acp' | 'serve' | 'plugins' | 'gh-bot' | 'provider' | 'mcp' | 'connect' | 'disconnect' | 'integrations' | 'completion' | 'stats' | 'export' | 'import' | 'pr';
+    subArgs: string[];
+    url?: string;
+    name?: string;
+    keyEnv?: string;
+    values: string[];
+    prompt?: string;
+    print: boolean;
+    model?: string;
+    reasoning?: ReasoningLevel;
+    lang?: Lang;
+    continue: boolean;
+    resume: boolean;
+    allowAll: boolean;
+    allowedTools: string[];
+    worktree?: string;
+    outputFormat?: 'text' | 'json' | 'stream-json';
+    permissionMode?: PermissionMode;
+    help: boolean;
+    version: boolean;
+    loginProvider?: string;
+    /** `bccli login` — pakai prompt API key lama, bukan OAuth. */
+    key: boolean;
+    /** `bccli login` — hapus credential tersimpan (hanya openrouter via OAuth path). */
+    logout: boolean;
+    /** `bccli serve` listener port (0 = ephemeral). */
+    port?: number;
+    /** `bccli serve` bind address — loopback only. */
+    host?: string;
+    /** `bccli serve` bearer token override. */
+    token?: string;
+    /** Base URL override for `bccli mcp serve`. */
+    baseUrl?: string;
+    /** `bccli stats` — only count sessions from the last N days (default 7). */
+    days?: number;
+    /** `bccli stats` - cap the tool-usage list to the top N tools. */
+    tools?: number;
+    /** `bccli stats [--models [N|all]]` - show per-model token usage (top N or all). */
+    models?: string;
+    /** `bccli stats --project <slug>` - filter rows; empty string = current project. */
+    project?: string;
+    /** `bccli pr` — explicit owner/name repo when cwd is not a git checkout of the PR. */
+    repo?: string;
+    /** Automatic fallback model ref when the primary model fails with 429 or 5xx. */
+    fallbackModel?: string;
+}
+/** Help text in the current language (call it after the language is set). */
+export declare const helpText: () => string;
+export declare function parseCliArgs(argv: string[]): CliArgs;

package/dist/budget.d.ts

@@ -0,0 +1,23 @@
+export interface UsageCap {
+    /** Total tokens (input + output) per session. */
+    tokens?: number;
+    /** Total USD per session; requires `prices`. */
+    usd?: number;
+    /** USD per 1M tokens, keyed by a substring of the model ref (first match wins). */
+    prices?: Record<string, {
+        input: number;
+        output: number;
+    }>;
+}
+export declare function priceFor(cap: UsageCap, model: string): {
+    input: number;
+    output: number;
+} | undefined;
+export declare function budgetStatus(usage: {
+    inputTokens: number;
+    outputTokens: number;
+}, cap: UsageCap | undefined, model: string): {
+    kind: 'tokens' | 'usd';
+    used: number;
+    limit: number;
+} | undefined;

package/dist/checkpoints.d.ts

@@ -0,0 +1,35 @@
+interface UndoResult {
+    restored: string[];
+    deleted: string[];
+    skipped: string[];
+    failed: string[];
+}
+export declare class CheckpointStore {
+    private readonly history;
+    private current;
+    private pendingRedo;
+    private offset;
+    beginTurn(): void;
+    /** Call before writing a file; only the first snapshot per file per turn is kept. */
+    snapshot(path: string): Promise<void>;
+    turns(): number;
+    /** Total entries including empty ones; used to align file snapshots with conversation turns. */
+    entries(): number;
+    undo(): Promise<UndoResult | undefined>;
+    /**
+     * Revert every snapshot taken at or after the `keep`-th entry, newest first, so the files
+     * end in the state from before the turn that created entry `keep`. Used by /rewind, which
+     * pairs conversation turns with the entry count recorded when each turn started.
+     */
+    undoTo(keep: number): Promise<UndoResult | undefined>;
+    /**
+     * Re-apply the file state captured right before the last undo/undoTo.
+     * Only file edits are redone, never conversation history. Invalidated by
+     * any new snapshot turn or clear(), so redo can't resurrect stale content.
+     */
+    redo(): Promise<UndoResult | undefined>;
+    /** Records the on-disk state of a turn's paths so redo() can put it back. */
+    private captureFor;
+    clear(): void;
+}
+export {};

package/dist/{chunk-6M5X2KZE.js → chunk-266SY3MC.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 
 // src/version.ts
-var VERSION = true ? "0.4.0" : "0.0.0-dev";
+var VERSION = true ? "0.4.2" : "0.0.0-dev";
 
 export {
   VERSION

package/dist/{chunk-7F3IWBPQ.js → chunk-3I5BW3ZT.js}

@@ -1,8 +1,12 @@
 #!/usr/bin/env node
+import {
+  quarantineCorruptFile,
+  writeFileAtomic
+} from "./chunk-RX4UQX43.js";
 
 // src/mcp/oauth.ts
 import { spawn } from "child_process";
-import { chmodSync, existsSync, mkdirSync, readFileSync, writeFileSync } from "fs";
+import { existsSync, mkdirSync, readFileSync } from "fs";
 import { createServer } from "http";
 import { dirname, join } from "path";
 import { auth } from "@modelcontextprotocol/sdk/client/auth.js";
@@ -12,16 +16,16 @@ function readStore(home) {
   if (!existsSync(path)) return {};
   try {
     return JSON.parse(readFileSync(path, "utf8"));
-  } catch {
+  } catch (error) {
+    quarantineCorruptFile(path, error);
     return {};
   }
 }
 function writeStore(home, store) {
   const path = oauthStorePath(home);
   mkdirSync(dirname(path), { recursive: true });
-  writeFileSync(path, `${JSON.stringify(store, null, 2)}
-`, { mode: 384 });
-  chmodSync(path, 384);
+  writeFileAtomic(path, `${JSON.stringify(store, null, 2)}
+`, 384);
 }
 function hasStoredTokens(home, serverUrl) {
   return Boolean(readStore(home)[serverUrl]?.tokens);
@@ -34,9 +38,15 @@ function clearTokens(home, serverUrl) {
   writeStore(home, store);
   return true;
 }
-function defaultOpen(url) {
-  const target = url.href;
-  const child = process.platform === "win32" ? spawn("cmd", ["/c", "start", "", target], { detached: true, stdio: "ignore" }) : spawn(process.platform === "darwin" ? "open" : "xdg-open", [target], { detached: true, stdio: "ignore" });
+function openCommand(url, platform) {
+  if (url.protocol !== "http:" && url.protocol !== "https:") return void 0;
+  if (platform === "win32") return { command: "rundll32.exe", args: ["url.dll,FileProtocolHandler", url.href] };
+  return { command: platform === "darwin" ? "open" : "xdg-open", args: [url.href] };
+}
+function defaultOpen(url, deps = {}) {
+  const plan = openCommand(url, deps.platform ?? process.platform);
+  if (!plan) return;
+  const child = (deps.spawn ?? spawn)(plan.command, plan.args, { detached: true, stdio: "ignore" });
   child.unref();
   child.on("error", () => {
   });
@@ -146,8 +156,9 @@ async function authorizeMcpServer(spec, deps) {
       res.end("<html><body><p>BCCLI is connected. You can close this tab.</p></body></html>");
       settle.resolve(code);
     } catch (error) {
+      const escaped = error.message.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;");
       res.writeHead(400, { "content-type": "text/html; charset=utf-8" });
-      res.end(`<html><body><p>${error.message}</p></body></html>`);
+      res.end(`<html><body><p>${escaped}</p></body></html>`);
       settle.reject(error);
     }
   });

package/dist/{chunk-UZWJQF67.js → chunk-3WHI4LN3.js}

@@ -458,7 +458,8 @@ Opsi:
   "gh is not available ({error}); install GitHub CLI first": "gh tidak tersedia ({error}); pasang GitHub CLI dulu",
   "gh pr checkout {n} failed (exit {code})": "gh pr checkout {n} gagal (exit {code})",
   "checked out PR #{n}": "checkout PR #{n} selesai",
-  "--allow-all: the PR task runs every tool without asking": "--allow-all: tugas PR menjalankan semua alat tanpa bertanya"
+  "--allow-all: the PR task runs every tool without asking": "--allow-all: tugas PR menjalankan semua alat tanpa bertanya",
+  "view or change configuration settings": "lihat atau ubah pengaturan konfigurasi"
 };
 
 // src/i18n.ts