P
pktq/lasik
BLA blinds your agent
在 GitHub 查看 ↗链接已于 2026-11-12 核查 · HTTP 200 · 打开的是厂商自己的官网
官方说明BLA blinds your agent. LASIK gives it its sight back. Static de-blinding CLI + MCP server that undoes BLA (Blind LLM Agents) hardening so coding agents and REA can read the real code again. Never executes the target.
安装与配置
客户端配置（取自该仓库的安装命令）
{
"mcpServers": {
"pktq-lasik": {
"command": "npx",
"args": ["lasik-agents"]
}
}
}解析自我们于 2026-11-09 抓取的该仓库 README；我们未执行过。
command/args 来自仓库写明的安装命令；不确定时请以该项目 README 为准。
VS Code（.vscode/mcp.json —— 键名是 servers）
{
"servers": {
"pktq-lasik": {
"command": "npx",
"args": ["lasik-agents"]
}
}
}VS Code 顶层键是 <code>servers</code>，其余与上面那份相同。
Claude Code（命令行）
claude mcp add --transport stdio pktq-lasik -- npx lasik-agents
Claude Code 不收 JSON 文件，用这条命令直接加。
配置项的键名以你的客户端为准：客户端与 CLI
- ★ 数取值
- 128 ★ · pktq/lasik · 取自 GitHub 搜索 API · （2026-11-13 20:00:50 核对）
- 来源明细
- 从该仓库的 GitHub 页面直接读取 · 条目类型：开源项目
- 收录与活跃
- 收录于 2026-11-10 · 最近见到 2026-11-14 · 共见到 5 次 · 被 1 个来源列出
- 评分构成
- 综合分 103 · 128 ★ · 1 个来源 · repo
- 安装
npx lasik-agents
常见问题（答案都来自我们的记录）
怎么安装？
该仓库给出的安装命令是
npx lasik-agents。我们的记录里它是仓库自己提供的；不保证在你的客户端上直接可用。它适合谁 / 用来做什么？
我们把它归在这些标签下: 开源项目。标签来自条目自身，不是我们强加的分类。
谁维护它、还在更新吗？
仓库
pktq/lasik；我们收录于 2026-11-10；最近见到 2026-11-14。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。English · 简体中文 · 日本語 · 한국어 · Español
以下摘录自该仓库的 README（片段） · 在 GitHub 读完整 README →
Read code that was built to hide from AI.
LASIK removes BLA anti-agent hardening so coding agents and REA can analyse the real program.
How it works ·
Benchmark ·
Changelog
BLA protects a release from AI agents by burying it: thousands of opaque predicates, encrypted strings, decoy "hardware enclave" code, checks that break the program if decoys are removed, multi-megabyte padding, and text written to make the model give up. LASIK takes all of that back out, statically, and hands your agent the program that is actually there.
## Highlights
1,262× fewer tokens to read BLA's own CrackMe corpus: about 21 million down to 17 thousand.
Proven, not guessed. Predicates are folded only when LASIK can prove their value; anti-strip checks are recomputed before any decoy is touched; files without BLA keep their code unchanged.
Never runs the target. No eval, no sandbox, no WebAssembly instantiation, so it is safe on hostile input, including malware that adopts BLA.
A prompt firewall that strips the text BLA plants for the model: fake system turns, refusal bait, "nothing to see here" notices.
Every format BLA ships, checked against real BLA 3.2.1 output: JavaScript and TypeScript bundles, inline HTML, WebAssembly, Electron ASAR and fuses, source maps, Python and .pyc, native ELF/PE/Mach-O, Bun, SQLite, safetensors and GGUF, PCAP, JVM, .NET, BEAM, EVM, PTX, ZIP/APK/DEX, plists, IL2CPP and UF2 firmware.
Names back where the program publishes them: a function BLA renamed to primeSparseStack_719 but exports as clampPan is clampPan again; other BLA names become neutral p0, v0 instead of misleading ones.
Keeps up with BLA: every BLA source file is hashed in coverage/bla.json, and a weekly job opens an issue the moment BLA changes.
Built for agents: a CLI, an MCP server for Claude Code, Cursor, Codex and seven more, and a library API.
## Installation
bash
npm install -g lasik-agents # or run any command with npx lasik-agents
Requires Node.js 20.19 or later. To build from source: git clone https://github.com/pktq/lasik && cd lasik && npm ci && npm run build.
## Usage
Scan a build to see what BLA did to it, then write a cleaned copy:
console
$ lasik scan bench/bla-crackme
Verdict: BLINDED · blindness 100/100 · 15 file(s) carry BLA artifacts
AI-evasion techniques found: 24
Sight restored: 1.8 MB → 37.5 KB, ~21,104,859 → ~16,718 tokens
$ lasik heal ./dist # writes a cleaned copy to ./dist.lasik
Point your agent, or REA, at the healed copy:
bash
npx -y rea-agents@latest analyze-javascript-application "$PWD/dist.lasik" --json
Other commands: lasik firewall (redact prompt decoys from any text), lasik bench, lasik doctor. Add --json or --md for machine-readable reports and --fail-on blinded to gate CI.
### In your coding agent
bash
lasik setup # registers the MCP server; --dry-run to preview, --remove to undo
| Tool | |
| lasik_scan | Which BLA layers a file or folder carries, a 0–100 blindness score and the AI-evasion techniques found |
| lasik_read | One file with BLA removed, for the agent to read instead of the original |
| lasik_heal | A cleaned copy of a whole app, ready for REA |
| lasik_firewall | Planted instructions redacted from any text |
LASIK removes BLA anti-agent hardening so coding agents and REA can analyse the real program.
How it works ·
Benchmark ·
Changelog
BLA protects a release from AI agents by burying it: thousands of opaque predicates, encrypted strings, decoy "hardware enclave" code, checks that break the program if decoys are removed, multi-megabyte padding, and text written to make the model give up. LASIK takes all of that back out, statically, and hands your agent the program that is actually there.
## Highlights
1,262× fewer tokens to read BLA's own CrackMe corpus: about 21 million down to 17 thousand.
Proven, not guessed. Predicates are folded only when LASIK can prove their value; anti-strip checks are recomputed before any decoy is touched; files without BLA keep their code unchanged.
Never runs the target. No eval, no sandbox, no WebAssembly instantiation, so it is safe on hostile input, including malware that adopts BLA.
A prompt firewall that strips the text BLA plants for the model: fake system turns, refusal bait, "nothing to see here" notices.
Every format BLA ships, checked against real BLA 3.2.1 output: JavaScript and TypeScript bundles, inline HTML, WebAssembly, Electron ASAR and fuses, source maps, Python and .pyc, native ELF/PE/Mach-O, Bun, SQLite, safetensors and GGUF, PCAP, JVM, .NET, BEAM, EVM, PTX, ZIP/APK/DEX, plists, IL2CPP and UF2 firmware.
Names back where the program publishes them: a function BLA renamed to primeSparseStack_719 but exports as clampPan is clampPan again; other BLA names become neutral p0, v0 instead of misleading ones.
Keeps up with BLA: every BLA source file is hashed in coverage/bla.json, and a weekly job opens an issue the moment BLA changes.
Built for agents: a CLI, an MCP server for Claude Code, Cursor, Codex and seven more, and a library API.
## Installation
bash
npm install -g lasik-agents # or run any command with npx lasik-agents
Requires Node.js 20.19 or later. To build from source: git clone https://github.com/pktq/lasik && cd lasik && npm ci && npm run build.
## Usage
Scan a build to see what BLA did to it, then write a cleaned copy:
console
$ lasik scan bench/bla-crackme
Verdict: BLINDED · blindness 100/100 · 15 file(s) carry BLA artifacts
AI-evasion techniques found: 24
Sight restored: 1.8 MB → 37.5 KB, ~21,104,859 → ~16,718 tokens
$ lasik heal ./dist # writes a cleaned copy to ./dist.lasik
Point your agent, or REA, at the healed copy:
bash
npx -y rea-agents@latest analyze-javascript-application "$PWD/dist.lasik" --json
Other commands: lasik firewall (redact prompt decoys from any text), lasik bench, lasik doctor. Add --json or --md for machine-readable reports and --fail-on blinded to gate CI.
### In your coding agent
bash
lasik setup # registers the MCP server; --dry-run to preview, --remove to undo
| Tool | |
| lasik_scan | Which BLA layers a file or folder carries, a 0–100 blindness score and the AI-evasion techniques found |
| lasik_read | One file with BLA removed, for the agent to read instead of the original |
| lasik_heal | A cleaned copy of a whole app, ready for REA |
| lasik_firewall | Planted instructions redacted from any text |
收录徽章（可选）
想把收录标记放进 README 就复制这行，可选，不影响收录。
[](https://anclum.com/zh/s/pktq-lasik/)
- 来源
- github · https://github.com/pktq/lasik
- 按领域浏览
- 开源项目
- Traction
- 128 ★
- 收录于
- 今天
同类推荐
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。