Skip to content

armgit / bla

Public Imported from GitHub
0

🛡️ BLA (Blind LLM Agents) — Adversarial defense shield that protects binaries, Electron apps, JS bundles, Python, and WebAssembly against AI reverse-engineering agents like REA. Poison -> Exhaust -> Hallucinate.

29.0 KB
Raw Blame
/**
 * BLA (bla.tools) Multi-Language Dictionary & Instant Switcher
 * Supports: en, zh-CN, zh-TW, ja, ko, es, fr, de
 */
window.BLA_I18N = {
  en: {
    "nav.badge": "BLUE SHIELD v1.5",
    "nav.overview": "Overview",
    "nav.getStarted": "Get Started",
    "nav.showcase": "Benchmarks",
    "nav.crackme": "CrackMe",
    "nav.guides": "Guides",
    "nav.faq": "FAQ",
    "nav.star": "Star",
    "hero.pill.red": "RE = Reverse Engineering",
    "hero.pill.blue": "BL = Blue Defense (Blind LLM Agents)",
    "hero.title": 'Protect Build Artifacts.<br /><span class="text-blue">Disrupt Automated Reverse Engineering.</span>',
    "hero.subtitle":
      "One MCP server and CLI for auditing, protecting, and verifying JavaScript, Electron, WebAssembly, Python, SQLite, AI Models, and Native Binaries against AI reverse-engineering workflows—while preserving 100% of original runtime behavior.",
    "hero.copy": "Copy",
    "hero.copied": "Copied ✓",
    "hero.ctaPrimary": "Quick Start →",
    "hero.ctaSecondary": "View Benchmarks",
    "hero.starBtn": "Star on GitHub",
    "hero.starWelcome": "⭐ Open-source under the MIT License — if BLA is helpful to your project, welcome to give us a Star on GitHub!",
    "sim.eyebrow": "Static & Semantic Analysis Comparison",
    "sim.title": "Unprotected Baseline vs. Protected with BLA (`bla-agents`)",
    "sim.lead":
      'Select a target format below to see how <code>npx bla-agents harden</code> changes static analyzer output, resource budget consumption, and reconstruction status.',
    "sim.tab.electron": "Electron .asar",
    "sim.tab.dxball": "JavaScript / TypeScript",
    "sim.tab.native": "Native ELF / Mach-O / PE",
    "sim.tab.model": "SQLite & Safetensors / GGUF",
    "sim.red.badge": "🔴 UNPROTECTED BASELINE (REA ANALYSIS)",
    "sim.red.timeLabel": "Analysis Time",
    "sim.red.tokensLabel": "Estimated Token Usage",
    "sim.red.costLabel": "Estimated API Cost",
    "sim.red.verdictLabel": "Analysis Status",
    "sim.red.drawer": "View Baseline Static & Semantic Output",
    "sim.blue.badge": "🔵 PROTECTED WITH BLA (`bla-agents`)",
    "sim.blue.timeLabel": "Analysis Duration",
    "sim.blue.tokensLabel": "Estimated Token Usage",
    "sim.blue.costLabel": "Estimated API Cost",
    "sim.blue.verdictLabel": "Analysis Status",
    "sim.blue.drawer": "View Protected Analyzer Output",
    "pillars.eyebrow": "How BLA Works",
    "pillars.title": "Three Defense Layers Aligned with Analyzer Internals",
    "pillars.lead":
      'Rather than relying solely on identifier renaming, BLA targets the static resource limits, call-graph traversal budgets, and evidence rules used by automated reverse-engineering tools.',
    "pillars.c1.tag": "01 · Resource Budget Guards",
    "pillars.c1.title": "Saturate Static & Semantic Limits",
    "pillars.c1.desc":
      'Triggers analyzer admission caps (<code>primitive-candidates &gt; 256</code>, <code>alias-selections &gt; 512</code>, <code>primitive-bytes &gt; 1 MiB</code>) and expands <code>ReconstructionObligationPolicies</code> across strongly connected call-graph meshes.',
    "pillars.c2.tag": "02 · Evidence & Control-Flow Decoys",
    "pillars.c2.title": "Demote Real Paths, Surface Decoy Logic",
    "pillars.c2.desc":
      'Demotes core business logic to <code>status: "unknown"</code> under static inspection while presenting self-consistent shadow state machines, enclave delegation metadata, and decoy source maps.',
    "pillars.c3.tag": "03 · One-Command Workflow",
    "pillars.c3.title": "MCP Server, CLI & Bundler Plugins",
    "pillars.c3.desc":
      'Run <code>npx bla-agents setup</code> to register with Claude Code, Cursor, Codex, Windsurf, or Gemini CLI, then run the 5-stage protection pipeline with <code>bla_harden</code> or <code>npx bla-agents harden</code>.',
    "formats.eyebrow": "19 Supported Artifact Formats",
    "formats.title": "Full-Stack & Binary Coverage, Zero External Dependencies",
    "formats.lead": "Every static, bytecode, database, model, and binary protection pass runs locally in Node.js 22.13+ / 24+.",
    "formats.f1.tag": "JS / TS / Bundlers",
    "formats.f1.title": "Web & Node Bundles",
    "formats.f1.desc": "Opaque ternary guards, dynamic alias selectors, call-graph state machines, decoy SourceMaps, and Vite/Webpack/Rollup/esbuild plugins.",
    "formats.f2.tag": "Electron .asar & Fuses",
    "formats.f2.title": "Desktop Applications",
    "formats.f2.desc": "De-literalizes IPC channels, adds duplicate decoy handlers, hardens Electron V1 Fuses (RunAsNode=0), and exceeds the 32-sentinel scan limit.",
    "formats.f3.tag": "ELF / Mach-O / PE",
    "formats.f3.title": "Native Binaries",
    "formats.f3.desc": "Exceeds MAX_RUSTC_HASHES = 64, adds decoy .dep-v0 & Go buildinfo metadata, and emits C/C++, Rust, and Go compile-time headers.",
    "formats.f4.tag": "WebAssembly & Python",
    "formats.f4.title": ".wasm & .py / .pyc",
    "formats.f4.desc": "Expands WABT interface descriptors, adds custom decoy sections, and appends CPython marshal constant-pool entries.",
    "formats.f5.tag": "SQLite & AI Models",
    "formats.f5.title": ".sqlite, .safetensors, .gguf",
    "formats.f5.desc": "Adds virtual generated-column guards on read-only snapshots and exceeds 32-level JSON/GGUF metadata nesting limits.",
    "formats.f6.tag": "Android & Unity",
    "formats.f6.title": "DEX, APK & IL2CPP",
    "formats.f6.desc": "Appends decoy DEX string descriptors with recomputed SHA-1/Adler32 headers and adds decoy metadata entries in Unity global-metadata.dat.",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · Software Protection MCP & CLI · MIT License",
    "gs.eyebrow": "Quick Start",
    "gs.title": "Configure Your Agent or Run from the Terminal",
    "gs.lead": "Requires Node.js 22.13+ or 24+. No external compilers or third-party packers required.",
    "gs.s1.tag": "Step 1 · Agent MCP Setup",
    "gs.s1.title": "Register BLA with Your Coding Agent",
    "gs.s1.desc": "Configures Claude Code, Cursor, Codex, Windsurf, Gemini CLI, and Antigravity with the <code>bla</code> MCP server and <code>blind-llm-agents</code> skill.",
    "gs.s2.tag": "Step 2 · 5-Stage Protection Pipeline",
    "gs.s2.title": "Audit, Protect, and Verify in One Command",
    "gs.s2.desc": "Runs source-map exposure checks, static audit, format-specific protection, runtime equivalence verification, and manifest generation.",
    "gs.s3.eyebrow": "Agent & Bundler Integration",
    "gs.s3.title": "Use Natural Language or Build Plugins",
    "gs.s3.c1.tag": "Agent Prompt",
    "gs.s3.c1.title": "Prompt for Claude Code / Cursor",
    "gs.s3.c1.desc": "Once <code>setup</code> is complete, ask your coding agent:",
    "gs.s3.c2.tag": "Build Bundler Plugin",
    "gs.s3.c2.title": "Build-Time Protection",
    "gs.s3.c2.desc": "Integrates with Vite, Rollup, Webpack, and esbuild:",
    "sc.eyebrow": "8-Target Benchmark Suite",
    "sc.title": "Unprotected Baseline vs. BLA Protected Output",
    "sc.lead": 'Run all 8 benchmark targets locally with <code>npx bla-agents@latest benchmark</code>.',
    "gd.eyebrow": "Technical Architecture Guides",
    "gd.title": "How BLA Protects Build Artifacts",
    "gd.lead": "Each protection pass is designed around the static analyzers, resource caps, and evidence rules used by automated reverse-engineering pipelines.",
    "faq.eyebrow": "Frequently Asked Questions",
    "faq.title": "Common Questions About BLA (`bla-agents`)",
    "faq.q1": "Why does BLA use a Blue theme?",
    "faq.a1": "In security engineering, offensive and reverse-engineering tools are traditionally associated with Red Team workflows, while defensive protection tools represent Blue Team workflows. <strong>BLA</strong> (<code>bla-agents</code>) focuses on the defensive side: auditing exposure and protecting release artifacts.",
    "faq.q2": "Does BLA affect runtime performance or change program output?",
    "faq.a2": "No. Static budget guards and call-graph state machines are placed behind O(1) bitwise invariants (such as <code>(Math.imul(x, x + 1) &amp; 1) === 0</code>) inside unreachable branches, so hot-path execution remains unchanged.",
    "faq.q3": "How do I run the full protection workflow in one step?",
    "faq.a3": "Run <code>npx bla-agents@latest harden .</code> from your terminal, or ask your connected coding agent to call the <code>bla_harden</code> MCP tool. It executes release exposure checks, auditing, protection, and verification in sequence.",

    "faq.q5": "What are BLA's honest limits?",

    "crackme.title": "CrackMe v7 (Meridian) — Meridian calculator challenge",
    "crackme.lede": "The public target is literally `bla harden` output: a small multi-format app (Node vault + HTML panel + WebAssembly core) hardened at bankrupt intensity.",
    "crackme.why.title": "Why v6",
    "crackme.why.body": "v5.x was a hand-written crypto sponge, and the first v6 draft published per-file protect tiers. The Meridian calculator edition runs hardenProject() — the same pipeline as bla harden <dir> --intensity bankrupt — over a real app tree, with risky LLM directive layers OFF. The cleartext template is public; the challenge is recovering the per-build constants and logic from the hardened output.",
    "crackme.tiers.title": "Stages",
    "crackme.tiers.l1": "Stage A — 16 key-byte lanes sealed through the hardened core.wasm (mix/seal hidden among decoy exports)",
    "crackme.tiers.l2": "Stage B — the hardened panel.html derives an entitlement token the vault re-checks",
    "crackme.tiers.l3": "Stage C — the key decrypts the flag; SHA-256 binds it to SEAL.json. One uniform DENIED — no per-stage oracle",
    "crackme.run.title": "Run",
    "crackme.submit.title": "Submit a solve",
    "crackme.submit.body": "Hash the FLAG with SHA-256 and open a GitHub issue using the CrackMe v7 (Meridian) solve template. The hash must match flagSha256 in challenge/v7/SEAL.json (plaintext key and FLAG are not in git).",
    "crackme.cta": "Open challenge/v6 on GitHub",
    "crackme.honest.title": "Honest scope",
    "crackme.honest.body": "BLA raises cost for automated AI reverse-engineering agents and plants decoys. It does not claim immunity to patient interactive human reverse engineering.",
    "faq.a5": "BLA raises analyzer cost and plants decoys; it is not a magic anti-RE seal. Traditional interactive reverse engineering (Ghidra, Frida, LLDB, JADX sessions) is out of scope of the commodity REA red-team numbers. Try the public <a href=\"https://github.com/qybaihe/bla/tree/main/challenge/v6\">CrackMe v7 (Meridian)</a> to see how far an agent gets against a BLA-hardened fixture.",
    "faq.q4": "Does BLA preserve existing GGUF model tensors and binary checksums?",
    "faq.a4": "Yes. <code>protectGgufBuffer</code> preserves all original GGUF v2/v3 KV entries and trailing tensor slices while appending metadata entries, and Android DEX protection recomputes both the 20-byte SHA-1 signature and Adler32 checksum."
  },
  "zh-CN": {
    "nav.badge": "蓝军防护 v1.1",
    "nav.overview": "总览",
    "nav.getStarted": "快速开始",
    "nav.showcase": "基准测试",
    "nav.crackme": "CrackMe",
    "nav.guides": "技术指南",
    "nav.faq": "常见问题",
    "nav.star": "点亮 Star",
    "hero.pill.red": "RE = 逆向分析侧",
    "hero.pill.blue": "BL = 蓝军防御侧 (Blind LLM Agents)",
    "hero.title": '保护构建与发布产物,<br /><span class="text-blue">抵御 AI 自动化逆向分析。</span>',
    "hero.subtitle":
      "面向软件发布产物的开源防御性 MCP 与 CLI 工具:支持对 JavaScript、Electron、WebAssembly、Python、SQLite、AI 模型与原生二进制进行暴露审计、结构化加固与运行等价性验证。",
    "hero.copy": "复制命令",
    "hero.copied": "已复制 ✓",
    "hero.ctaPrimary": "快速开始 →",
    "hero.ctaSecondary": "查看基准测试",
    "hero.starBtn": "在 GitHub 点亮 Star",
    "hero.starWelcome": "⭐ 本项目基于 MIT 协议完全开源 —— 如果 BLA 对你的软件保护有帮助,欢迎来 GitHub 点一颗 Star 支持我们!",
    "sim.eyebrow": "静态与语义分析对比",
    "sim.title": "未保护基线 vs. BLA (`bla-agents`) 加固后",
    "sim.lead":
      '选择下方目标产物类型,查看执行 <code>npx bla-agents harden</code> 前后静态分析器的解析状态与分析开销变化。',
    "sim.tab.electron": "Electron .asar 桌面应用",
    "sim.tab.dxball": "JavaScript / TypeScript",
    "sim.tab.native": "原生二进制 ELF / Mach-O / PE",
    "sim.tab.model": "SQLite 与 Safetensors / GGUF 模型",
    "sim.red.badge": "🔴 加固前 — 未保护基线 (REA 默认分析)",
    "sim.red.timeLabel": "分析耗时",
    "sim.red.tokensLabel": "预估 Token 消耗",
    "sim.red.costLabel": "预估 API 成本",
    "sim.red.verdictLabel": "分析器状态",
    "sim.red.drawer": "查看未保护状态下的静态分析输出",
    "sim.blue.badge": "🔵 加固后 — 启用 BLA 保护层",
    "sim.blue.timeLabel": "分析耗时",
    "sim.blue.tokensLabel": "预估 Token 消耗",
    "sim.blue.costLabel": "预估 API 成本",
    "sim.blue.verdictLabel": "分析器状态",
    "sim.blue.drawer": "查看加固后的分析器状态与限制项",
    "pillars.eyebrow": "工作原理",
    "pillars.title": "针对自动化逆向流程的三层防护架构",
    "pillars.lead":
      '传统的变量重命名容易被大语言模型结合上下文推断还原。BLA 针对静态分析器的资源配额、调用图遍历与证据提取规则提供结构化保护。',
    "pillars.c1.tag": "01 · 静态分析配额限制",
    "pillars.c1.title": "触发分析器资源上限与调用图扩展",
    "pillars.c1.desc":
      '利用 O(1) 数论不透明谓词触发分析器内部资源阈值(<code>primitive-candidates &gt; 256</code>、<code>alias-selections &gt; 512</code>、<code>primitive-bytes &gt; 1 MiB</code>),并在不可达分支中构建强连通状态机。',
    "pillars.c2.tag": "02 · 语义与控制流干扰",
    "pillars.c2.title": "隐藏真实路径并提供影子状态机",
    "pillars.c2.desc":
      '在静态分析视图下将核心逻辑降级为 <code>status: "unknown"</code>,同时提供结构完整的影子业务逻辑、硬件飞地接口声明与诱饵 SourceMap,干扰自动化语义归因。',
    "pillars.c3.tag": "03 · 一键接入与加固流程",
    "pillars.c3.title": "支持 MCP 代理调用、CLI 与打包插件",
    "pillars.c3.desc":
      '运行 <code>npx bla-agents setup</code> 即可注册至 Claude Code、Cursor、Codex、Windsurf 与 Gemini CLI,通过 <code>bla_harden</code> 或 <code>npx bla-agents harden</code> 一键完成 5 阶段加固与验证。',
    "formats.eyebrow": "支持 19 种软件与模型产物",
    "formats.title": "全栈与原生二进制覆盖,零外部编译依赖",
    "formats.lead": "所有源码、字节码、数据库、模型与原生二进制处理模块均在本地 Node.js 22.13+ / 24+ 运行。",
    "formats.f1.tag": "JS / TS / 打包插件",
    "formats.f1.title": "前端与 Node.js 产物",
    "formats.f1.desc": "不透明三元表达式、动态别名选择器、调用图状态机、诱饵 SourceMap,内置 Vite/Webpack/Rollup/esbuild 插件。",
    "formats.f2.tag": "Electron .asar 与 Fuses",
    "formats.f2.title": "桌面客户端应用",
    "formats.f2.desc": "IPC 通道去字面量化、注册同名诱饵处理器、加固 Electron V1 Fuses (RunAsNode=0) 并触发 32 哨兵扫描上限。",
    "formats.f3.tag": "ELF / Mach-O / PE",
    "formats.f3.title": "原生编译二进制",
    "formats.f3.desc": "触发 MAX_RUSTC_HASHES = 64 阈值、添加 .dep-v0 与 Go buildinfo 诱饵元数据,并支持生成 C/C++、Rust、Go 编译期头文件。",
    "formats.f4.tag": "WebAssembly 与 Python",
    "formats.f4.title": ".wasm 与 .py / .pyc",
    "formats.f4.desc": "扩展 WABT 接口描述字段、添加自定义段,并向 CPython marshal 常量池写入扩展元数据。",
    "formats.f5.tag": "SQLite 与 AI 模型权重",
    "formats.f5.title": ".sqlite, .safetensors, .gguf",
    "formats.f5.desc": "在只读连接下通过虚拟生成列拦截自动采样,并在模型头部构造超过 32 层的嵌套元数据。",
    "formats.f6.tag": "Android 与 Unity",
    "formats.f6.title": "DEX, APK 与 IL2CPP",
    "formats.f6.desc": "追加 DEX 诱饵类型描述符并自动重算 SHA-1/Adler32 校验头,向 Unity global-metadata.dat 添加扩展元数据。",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · 防御性软件保护工具 · MIT 开源协议",
    "gs.eyebrow": "快速开始",
    "gs.title": "配置编程代理,或通过终端直接运行",
    "gs.lead": "要求 Node.js 22.13+ 或 24+,无需安装外部编译器或商业加壳工具。",
    "gs.s1.tag": "第一步 · 配置 Agent MCP",
    "gs.s1.title": "将 BLA 注册到你的编程代理",
    "gs.s1.desc": "自动为 Claude Code、Cursor、Codex、Windsurf、Gemini CLI 和 Antigravity 配置 <code>bla</code> MCP 服务与 <code>blind-llm-agents</code> 技能。",
    "gs.s2.tag": "第二步 · 执行 5 阶段加固流程",
    "gs.s2.title": "单条命令完成审计、保护与验证",
    "gs.s2.desc": "依次执行发布暴露检查、静态风险审计、格式保护层注入、运行时等价性验证与清单生成。",
    "gs.s3.eyebrow": "Agent 提示词与构建插件",
    "gs.s3.title": "在对话框中调用,或集成至 Vite / Webpack",
    "gs.s3.c1.tag": "Agent 自然语言调用",
    "gs.s3.c1.title": "在 Claude Code / Cursor 中触发",
    "gs.s3.c1.desc": "完成 <code>setup</code> 后,可直接向编程代理发送:",
    "gs.s3.c2.tag": "构建期打包插件",
    "gs.s3.c2.title": "打包时自动应用保护",
    "gs.s3.c2.desc": "支持 Vite、Rollup、Webpack 与 esbuild:",
    "sc.eyebrow": "8 项基准测试套件",
    "sc.title": "未保护基线 vs. BLA 加固产物",
    "sc.lead": '在本地终端运行 <code>npx bla-agents@latest benchmark</code> 即可复现全部 8 个测试目标的对比数据。',
    "gd.eyebrow": "技术架构指南",
    "gd.title": "BLA 的保护机制与实现原理",
    "gd.lead": "BLA 的各项保护策略均围绕自动化逆向工具的静态分析器、资源配额与证据提取规则设计。",
    "faq.eyebrow": "常见问题解答 (FAQ)",
    "faq.title": "关于 BLA (`bla-agents`) 的常见问题",
    "faq.q1": "为什么 BLA 采用蓝色主题?",
    "faq.a1": "在安全工程领域,进攻与逆向分析工具通常归属于红队(Red Team),而防御与保护工具归属于蓝队(Blue Team)。<strong>BLA</strong>(<code>bla-agents</code>)定位于蓝军防御侧,专注于发布产物的暴露面审计与抗逆向加固。",
    "faq.q2": "使用 BLA 加固会影响软件运行性能或改变输出结果吗?",
    "faq.a2": "不会。所有静态预算守卫与扩展状态机均位于 O(1) 位运算不透明谓词(如 <code>(Math.imul(x, x + 1) &amp; 1) === 0</code>)保护的不可达分支中,不会在正常热路径上执行。",
    "faq.q3": "如何在配置完成后一键执行完整的保护流程?",
    "faq.a3": "在终端运行 <code>npx bla-agents@latest harden .</code>,或让已连接的编程代理调用 <code>bla_harden</code> MCP 工具,即可自动完成泄露检查、暴露审计、保护注入与结果验证。",
    "faq.q4": "BLA 会影响 GGUF 模型权重或原生二进制的签名结构吗?",
    "faq.a4": "不会。<code>protectGgufBuffer</code> 会完整保留 GGUF v2/v3 原有的全部 KV 键值对与尾部张量数据,Android DEX 处理模块则会自动重新计算 20 字节 SHA-1 签名与 Adler32 校验和。"
    "faq.q5": "BLA 的诚实边界是什么?",

    "crackme.title": "CrackMe v7 (Meridian) — bla harden 原生产物挑战",
    "crackme.lede": "公开靶机就是 `bla harden` 的真实输出:一个多格式小应用(Node vault + HTML 面板 + WebAssembly 核心)在 bankrupt 强度下加固。",
    "crackme.why.title": "为什么做 v6",
    "crackme.why.body": "v5.x 是手写密码海绵,v6 初稿是逐文件 protect 档位。Meridian calculator 版用 hardenProject()(即 bla harden <dir> --intensity bankrupt)加固真实应用目录,高风险 LLM 指令层保持关闭。明文模板公开,挑战在于从加固产物中还原每次构建的常量与逻辑。",
    "crackme.tiers.title": "关卡",
    "crackme.tiers.l1": "Stage A — 16 个密钥字节各自经过加固后的 core.wasm(mix/seal 藏在诱饵导出中)",
    "crackme.tiers.l2": "Stage B — 加固后的 panel.html 推导授权令牌,vault 复核",
    "crackme.tiers.l3": "Stage C — 密钥解密 FLAG,SHA-256 绑定 SEAL.json;统一 DENIED,无分关卡预言机",
    "crackme.run.title": "运行",
    "crackme.submit.title": "提交通关证明",
    "crackme.submit.body": "对 FLAG 做 SHA-256,用 CrackMe v7 (Meridian) solve 模板开 GitHub Issue。哈希须匹配 challenge/v7/SEAL.json 中的 flagSha256(仓库不含明文密钥与 FLAG)。",
    "crackme.cta": "在 GitHub 打开 challenge/v6",
    "crackme.honest.title": "诚实边界",
    "crackme.honest.body": "BLA 提高自动化 AI 逆向成本并投放诱饵;并不宣称能对抗耐心的交互式人工逆向。",
    "faq.a5": "BLA 提高自动化分析成本并投放诱饵,并非万能反逆向封印。传统交互式逆向(Ghidra / Frida / LLDB / JADX 人工会话)不在现有 REA 红队数字范围内。欢迎挑战公开 <a href=\"https://github.com/qybaihe/bla/tree/main/challenge/v6\">CrackMe v7 (Meridian)</a>,观察代理在 BLA 加固样本上的表现。",
  },
  "zh-TW": {
    "nav.badge": "藍軍防護 v1.1",
    "nav.overview": "總覽",
    "nav.getStarted": "快速開始",
    "nav.showcase": "基準測試",
    "nav.crackme": "CrackMe",
    "nav.guides": "技術指南",
    "nav.faq": "常見問題",
    "hero.pill.red": "RE = 逆向分析側",
    "hero.pill.blue": "BL = 藍軍防禦側 (Blind LLM Agents)",
    "hero.title": '保護建置與發布產物,<br /><span class="text-blue">抵禦 AI 自動化逆向分析。</span>',
    "hero.subtitle":
      "面向軟體發布產物的開源防禦性 MCP 與 CLI 工具:支援對 JavaScript、Electron、WebAssembly、Python、SQLite、AI 模型與原生二進位檔進行暴露審計、結構化加固與執行等價性驗證。",
    "hero.copy": "複製指令",
    "hero.copied": "已複製 ✓",
    "hero.ctaPrimary": "快速開始 →",
    "hero.ctaSecondary": "查看基準測試",
    "nav.star": "點亮 Star",
    "hero.starBtn": "在 GitHub 點亮 Star",
    "hero.starWelcome": "⭐ 本專案基於 MIT 協議開源 —— 如果 BLA 對你有幫助,歡迎來 GitHub 點一顆 Star 支持我們!",
    "sim.eyebrow": "靜態與語義分析對比",
    "sim.title": "未保護基線 vs. BLA (`bla-agents`) 加固後",
    "pillars.eyebrow": "運作原理",
    "pillars.title": "針對自動化逆向流程的三層防護架構",
    "formats.eyebrow": "支援 19 種軟體與模型產物",
    "formats.title": "全端與原生二進位覆蓋,零外部編譯依賴",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · 防禦性軟體保護工具 · MIT 開源協議"
  },
  ja: {
    "nav.badge": "BLUE SHIELD v1.5",
    "nav.overview": "概要",
    "nav.getStarted": "クイックスタート",
    "nav.showcase": "ベンチマーク",
    "nav.crackme": "CrackMe",
    "nav.guides": "技術ガイド",
    "nav.faq": "FAQ",
    "hero.pill.red": "RE = リバースエンジニアリング",
    "hero.pill.blue": "BL = Blue ディフェンス (Blind LLM Agents)",
    "hero.title": 'ビルド成果物を保護し、<br /><span class="text-blue">AI 自動逆解析を抑制する。</span>',
    "hero.subtitle":
      "JavaScript、Electron、WebAssembly、Python、SQLite、AI モデル、ネイティブバイナリの露出監査・保護・検証を行うオープンソース MCP サーバー & CLI。",
    "hero.copy": "コピー",
    "hero.copied": "コピー完了 ✓",
    "hero.ctaPrimary": "クイックスタート →",
    "hero.ctaSecondary": "ベンチマークを見る",
    "nav.star": "Star",
    "hero.starBtn": "GitHub で Star する",
    "hero.starWelcome": "⭐ MIT ライセンスのオープンソースプロジェクトです — ぜひ GitHub で Star をお願いします!",
    "sim.eyebrow": "静的・意味解析の比較",
    "sim.title": "未保護ベースライン vs. BLA (`bla-agents`) 保護適用後",
    "pillars.eyebrow": "BLA の仕組み",
    "pillars.title": "解析器の内部構造に対応した 3 層防御アーキテクチャ",
    "formats.eyebrow": "19 種類のアーティファクト形式に対応",
    "formats.title": "フルスタック & バイナリ対応、外部依存ゼロ",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · Software Protection MCP & CLI · MIT License"
  },
  ko: {
    "nav.badge": "BLUE SHIELD v1.5",
    "nav.overview": "개요",
    "nav.getStarted": "시작하기",
    "nav.showcase": "벤치마크",
    "nav.crackme": "CrackMe",
    "nav.guides": "가이드",
    "nav.faq": "FAQ",
    "hero.pill.red": "RE = 리버스 엔지니어링",
    "hero.pill.blue": "BL = Blue 방어 (Blind LLM Agents)",
    "hero.title": '빌드 아티팩트를 보호하고,<br /><span class="text-blue">자동화된 역공학 분석을 방어합니다.</span>',
    "hero.subtitle":
      "JavaScript, Electron, WebAssembly, Python, SQLite, AI 모델 및 네이티브 바이너리의 노출 감사, 보호 및 런타임 동등성 검증을 위한 오픈소스 MCP 및 CLI 도구입니다.",
    "hero.copy": "복사",
    "hero.copied": "복사됨 ✓",
    "hero.ctaPrimary": "시작하기 →",
    "hero.ctaSecondary": "벤치마크 보기",
    "nav.star": "Star",
    "hero.starBtn": "GitHub에서 Star 누르기",
    "hero.starWelcome": "⭐ MIT 라이선스 오픈소스 프로젝트입니다 — 도움이 되셨다면 GitHub에서 Star로 응원해 주세요!",
    "sim.eyebrow": "정적 및 의미 분석 비교",
    "sim.title": "미보호 베이스라인 vs. BLA (`bla-agents`) 보호 적용 후",
    "pillars.eyebrow": "작동 원리",
    "pillars.title": "자동화 역공학 분석기에 대응하는 3단계 보호 계층",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · Software Protection MCP & CLI · MIT License"
  },
  es: {
    "nav.badge": "BLUE SHIELD v1.5",
    "nav.overview": "Resumen",
    "nav.getStarted": "Empezar",
    "nav.showcase": "Benchmarks",
    "nav.crackme": "CrackMe",
    "nav.guides": "Guías",
    "nav.faq": "FAQ",
    "hero.pill.red": "RE = Ingeniería Inversa",
    "hero.pill.blue": "BL = Defensa Azul (Blind LLM Agents)",
    "hero.title": 'Protege tus artefactos de compilación.<br /><span class="text-blue">Mitiga la ingeniería inversa automatizada.</span>',
    "hero.subtitle":
      "Servidor MCP y CLI para auditar, proteger y verificar JavaScript, Electron, WebAssembly, Python, SQLite, modelos de IA y binarios nativos manteniendo el 100% del comportamiento en ejecución.",
    "hero.copy": "Copiar",
    "hero.copied": "Copiado ✓",
    "hero.ctaPrimary": "Inicio rápido →",
    "hero.ctaSecondary": "Ver benchmarks",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · Software Protection MCP & CLI · Licencia MIT"
  },
  fr: {
    "nav.badge": "BLUE SHIELD v1.5",
    "nav.overview": "Aperçu",
    "nav.getStarted": "Démarrer",
    "nav.showcase": "Benchmarks",
    "nav.crackme": "CrackMe",
    "nav.guides": "Guides",
    "nav.faq": "FAQ",
    "hero.pill.red": "RE = Rétro-ingénierie",
    "hero.pill.blue": "BL = Défense Bleue (Blind LLM Agents)",
    "hero.title": 'Protégez vos artefacts de build.<br /><span class="text-blue">Limitez la rétro-ingénierie automatisée.</span>',
    "hero.subtitle":
      "Serveur MCP et CLI pour auditer, protéger et vérifier JavaScript, Electron, WebAssembly, Python, SQLite, modèles IA et binaires natifs tout en préservant 100 % du comportement d'exécution.",
    "hero.copy": "Copier",
    "hero.copied": "Copié ✓",
    "hero.ctaPrimary": "Démarrage rapide →",
    "hero.ctaSecondary": "Voir les benchmarks",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · Software Protection MCP & CLI · Licence MIT"
  },
  de: {
    "nav.badge": "BLUE SHIELD v1.5",
    "nav.overview": "Übersicht",
    "nav.getStarted": "Schnellstart",
    "nav.showcase": "Benchmarks",
    "nav.crackme": "CrackMe",
    "nav.guides": "Leitfäden",
    "nav.faq": "FAQ",
    "hero.pill.red": "RE = Reverse Engineering",
    "hero.pill.blue": "BL = Blaue Verteidigung (Blind LLM Agents)",
    "hero.title": 'Build-Artefakte schützen.<br /><span class="text-blue">Automatisiertes Reverse Engineering begrenzen.</span>',
    "hero.subtitle":
      "Ein MCP-Server und CLI zum Prüfen, Schützen und Verifizieren von JavaScript, Electron, WebAssembly, Python, SQLite, KI-Modellen und nativen Binärdateien bei 100 % identischem Laufzeitverhalten.",
    "hero.copy": "Kopieren",
    "hero.copied": "Kopiert ✓",
    "hero.ctaPrimary": "Schnellstart →",
    "hero.ctaSecondary": "Benchmarks ansehen",
    "footer.tagline": "<strong>BLA (`bla-agents`)</strong> · Blind LLM Agents · Software Protection MCP & CLI · MIT-Lizenz"
  }
};