About Progmune 关于我们
English below ｜ 中文见下English
Progmune is an AI Trust Decision Engine for AI-generated software. It verifies that code — written by AI or humans — follows correct protocol lifecycles: TLS handshakes, authentication flows, payment integrity, and resource management. These violations span sequences of function calls, which makes them invisible to pattern-based SAST and dependency-level SCA tools.
Progmune works through two complementary paths: a generation-time intercept (code produced through Progmune is validated before it ever lands on disk) and a post-hoc verify path (any codebase can be checked via CLI, SDK, MCP, or CI). Verdicts come from a deterministic protocol state machine — not from an LLM judging itself. Every result ships with a Trust Score, a Decision (APPROVED / NEEDS_REVIEW / BLOCKED), and an auditable evidence chain.
The project is open source (MIT), published on npm and GitHub. It was founded in 2026 by a solo maintainer. Benchmarks are published openly: TypeScript 795-gold blind test at 98.5% recall / 100% precision / 0 FP; Python at 100% / 100% / 0 FP. Three independent third-party audits (2026-09) were completed and all findings addressed; the response is public in the repository.
Progmune positions itself honestly as an auxiliary checker — evidence for human review — not a deployment gate. Questions or feedback: shenlian1983@qq.com.
中文
Progmune 是面向 AI 生成软件的 AI 信任决策引擎。它验证代码——无论由 AI 还是人类编写——是否遵循正确的协议生命周期：TLS 握手、认证流、支付完整性、资源管理。这类违规跨越函数调用序列，模式级 SAST 与依赖级 SCA 都看不见。
Progmune 通过两条互补路径工作：生成时拦截（经由 Progmune 产出的代码在落盘前完成校验）与事后验证（任何代码库都可通过 CLI、SDK、MCP 或 CI 检查）。判定来自确定性协议状态机——不是 LLM 自审自判。每次结果附带 Trust Score、Decision（APPROVED / NEEDS_REVIEW / BLOCKED）与可审计的证据链。
项目为开源（MIT），发布于 npm 与 GitHub，2026 年由独立维护者创立。基准公开：TypeScript 795 金标盲测 Recall 98.5% / Precision 100% / 0 误报；Python 100% / 100% / 0 误报。三份独立第三方审计（2026-09）已全部回应并修复，回应文档在仓库中公开。
Progmune 如实定位为辅助检查器——为人工审查提供证据——而非部署门禁。问题或反馈：shenlian1983@qq.com。