Q
qira-dev/ShellMind-Guard
Compiler-grade multi-file security, business-logic & concurrency verifier in Rust.
更新 2026-11-19
在 GitHub 查看 ↗链接已于 2026-11-17 核查 · HTTP 200 · 打开的是厂商自己的官网
- ★ 数取值
- 0 ★ · qira-dev/ShellMind-Guard · 取自 GitHub 搜索 API · (2026-11-19 20:00:48 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 2026-11-16 · 最近见到 2026-11-20 · 共见到 5 次 · 被 2 个来源列出
- 评分构成
- 综合分 203 · 2 个来源 · repo
常见问题(答案都来自我们的记录)
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
qira-dev/ShellMind-Guard;我们收录于 2026-11-16;最近见到 2026-11-20。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。README 原文(英文,未翻译) · 在 GitHub 读完整 README →
# ShellMind Guard
以下摘录自该仓库的 README(片段)
Compiler-Grade Multi-File Security, Business-Logic & Concurrency Verifier
Traces cross-file taint and guard invariants via an $O(V + E)$ Guard-Set Lattice across 16 languages (TypeScript, JavaScript, Go, Python, Rust, PHP, Java, C#, Kotlin, Swift, Dart, C, C++, Ruby, Bash/Shell, and Terraform/HCL)—then mathematically proves exploitability inside an unprivileged Linux Bubblewrap (bwrap) sandbox with zero AI API cost.
Architecture · 6-Layer Engine · Quickstart · AI Agent MCP · Benchmarks · Blueprint
## Why ShellMind Guard?
Modern software—especially high-velocity, AI-assisted full-stack codebases—suffers from vulnerabilities that span multiple files and execution layers: a route in routes/checkout.ts delegates to services/ledger.ts, which mutates a balance in db/queries.ts without an authentication middleware, ownership filter, or SELECT ... FOR UPDATE transaction lock.
Conventional security tools fail on these bugs in two predictable ways:
1. Regex / Single-File SAST Scanners lack cross-file call-graph and middleware-order semantics, burying engineering teams under hundreds of false positives while missing multi-hop business-logic and race-condition flaws.
2. LLM-Only Code Scanners burn tens of thousands of tokens per scan, hallucinate findings on sanitized paths, cannot execute deterministic data-flow proofs, and cannot verify whether a concurrency bug is actually exploitable at runtime.
ShellMind Guard solves both problems with a deterministic, systems-engineered architecture:
Two-Phase Incremental Ingestion (BLAKE3 + SQLite WAL): Fingerprints 13,000+ files at >1 GB/s, caching per-file Tree-sitter AST summaries in SQLite while always re-linking the global cross-file call graph in RAM ( /state.db)"]
B1 --> B2 --> B3
end
subgraph S2["Stage 2: Polyglot AST & O(V+E) Guard-Set Lattice (sh-graph)"]
C1["Rayon Parallel Tree-sitter AST (16 Languages)\nTS/JS · Go · Python · Rust · PHP · Java · C# · Kotlin · Swift · Dart · C/C++ · Ruby · Bash · HCL"]
C2["Cross-File Module, Header & IaC Linker\ntsconfig paths · barrels · Go/Java/Kotlin/C#/PHP/Dart/Ruby/Bash/C++ headers · Terraform modules"]
C3["6-Layer Guard-Set Lattice & Sink-Typed Bitmask Taint\nG(v) = ⋂ (G(p) ∪ EdgeGuards(p → v))"]
C1 --> C2 --> C3
end
subgraph S3["Stage 3: OS-Isolated Runtime Prover (sh-sandbox)"]
D1["Linux Bubblewrap (/usr/bin/bwrap)\n--unshare-all · --ro-bind · --tmpfs · PGID Reaper"]
D2["12-Way Barrier-Synchronized Prober\nAuth Bypass, Cross-Tenant IDOR & Last-Byte Sync Race"]
D1 --> D2
end
subgraph S4["Stage 4: Actionable Outputs (sh-report / sh-ai)"]
E1["SARIF 2.1.0 + Confidence\nGitHub Code Scanning"]
E2["Interactive HTML/SVG\nMulti-File Witness Graph"]
E3["Minimal Patch Prompt ( S1 --> S2
S2 -->|"Static Findings"| S4
S2 -->|"--boot-cmd"| S3
S3 -->|"PROVEN / MITIGATED"| S4
### Mathematical Formulation: The $O(V + E)$ Guard-Set Lattice
Instead of enumerating all simple paths between routes and sinks (which degrades to $O(V!)$ on cyclic call graphs), ShellMind Guard formulates security invariant verification as a forward data-flow monotone lattice over the finite universe of security guards $\mathcal{U} = \{\text{Auth}, \text{TxLock}, \text{OwnershipCheck}, \text{SqlSanitizer}, \text{CmdSanitizer}, \text{PathSanitizer}, \text{SsrfSanitizer}, \text{XssSanitizer}, \text{CsrfProtection}\}$ paired with a sink-typed bitmask taint semilattice (preventing cross-sink sanitizer confusion) and explicit finding confidence grading (PROVEN, CONFIRMED, HEURISTIC):
Traces cross-file taint and guard invariants via an $O(V + E)$ Guard-Set Lattice across 16 languages (TypeScript, JavaScript, Go, Python, Rust, PHP, Java, C#, Kotlin, Swift, Dart, C, C++, Ruby, Bash/Shell, and Terraform/HCL)—then mathematically proves exploitability inside an unprivileged Linux Bubblewrap (bwrap) sandbox with zero AI API cost.
Architecture · 6-Layer Engine · Quickstart · AI Agent MCP · Benchmarks · Blueprint
## Why ShellMind Guard?
Modern software—especially high-velocity, AI-assisted full-stack codebases—suffers from vulnerabilities that span multiple files and execution layers: a route in routes/checkout.ts delegates to services/ledger.ts, which mutates a balance in db/queries.ts without an authentication middleware, ownership filter, or SELECT ... FOR UPDATE transaction lock.
Conventional security tools fail on these bugs in two predictable ways:
1. Regex / Single-File SAST Scanners lack cross-file call-graph and middleware-order semantics, burying engineering teams under hundreds of false positives while missing multi-hop business-logic and race-condition flaws.
2. LLM-Only Code Scanners burn tens of thousands of tokens per scan, hallucinate findings on sanitized paths, cannot execute deterministic data-flow proofs, and cannot verify whether a concurrency bug is actually exploitable at runtime.
ShellMind Guard solves both problems with a deterministic, systems-engineered architecture:
Two-Phase Incremental Ingestion (BLAKE3 + SQLite WAL): Fingerprints 13,000+ files at >1 GB/s, caching per-file Tree-sitter AST summaries in SQLite while always re-linking the global cross-file call graph in RAM ( /state.db)"]
B1 --> B2 --> B3
end
subgraph S2["Stage 2: Polyglot AST & O(V+E) Guard-Set Lattice (sh-graph)"]
C1["Rayon Parallel Tree-sitter AST (16 Languages)\nTS/JS · Go · Python · Rust · PHP · Java · C# · Kotlin · Swift · Dart · C/C++ · Ruby · Bash · HCL"]
C2["Cross-File Module, Header & IaC Linker\ntsconfig paths · barrels · Go/Java/Kotlin/C#/PHP/Dart/Ruby/Bash/C++ headers · Terraform modules"]
C3["6-Layer Guard-Set Lattice & Sink-Typed Bitmask Taint\nG(v) = ⋂ (G(p) ∪ EdgeGuards(p → v))"]
C1 --> C2 --> C3
end
subgraph S3["Stage 3: OS-Isolated Runtime Prover (sh-sandbox)"]
D1["Linux Bubblewrap (/usr/bin/bwrap)\n--unshare-all · --ro-bind · --tmpfs · PGID Reaper"]
D2["12-Way Barrier-Synchronized Prober\nAuth Bypass, Cross-Tenant IDOR & Last-Byte Sync Race"]
D1 --> D2
end
subgraph S4["Stage 4: Actionable Outputs (sh-report / sh-ai)"]
E1["SARIF 2.1.0 + Confidence\nGitHub Code Scanning"]
E2["Interactive HTML/SVG\nMulti-File Witness Graph"]
E3["Minimal Patch Prompt ( S1 --> S2
S2 -->|"Static Findings"| S4
S2 -->|"--boot-cmd"| S3
S3 -->|"PROVEN / MITIGATED"| S4
### Mathematical Formulation: The $O(V + E)$ Guard-Set Lattice
Instead of enumerating all simple paths between routes and sinks (which degrades to $O(V!)$ on cyclic call graphs), ShellMind Guard formulates security invariant verification as a forward data-flow monotone lattice over the finite universe of security guards $\mathcal{U} = \{\text{Auth}, \text{TxLock}, \text{OwnershipCheck}, \text{SqlSanitizer}, \text{CmdSanitizer}, \text{PathSanitizer}, \text{SsrfSanitizer}, \text{XssSanitizer}, \text{CsrfProtection}\}$ paired with a sink-typed bitmask taint semilattice (preventing cross-sink sanitizer confusion) and explicit finding confidence grading (PROVEN, CONFIRMED, HEURISTIC):
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/qira-dev-shellmind-guard/)
- 来源
- github · https://github.com/qira-dev/ShellMind-Guard
- 按领域浏览
- 开源项目
- 收录来源
- 2 个独立来源
- 收录于
- 今天
更多条目
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。