ai-security-scanner - Desktop app and Agent Skills that run established

项目说明

ai-security-scanner

Project website · 繁體中文 · Documentation · Releases

Security checks across repositories, websites, and internal systems, through a desktop app or an Agent Skill for Claude Code and Codex. Select the targets, confirm authorization for network checks, and start one scan. Established upstream scanners provide the detection; thin adapters and output con bring their results into one standardized, prioritized report.

Start here

Download the v0.4.0 stable release for your computer:

| Computer | Installer | First launch |
| --- | --- | --- |
| macOS, Apple silicon or Intel | ai-security-scanner_0.4.0_uni.dmg | Drag the app into Applications. It is not notarized, so run xattr -dr com.apple.quarantine /Applications/ai-security-scanner.app once in Terminal before opening it. |
| Windows x86-64 | ai-security-scanner_0.4.0_x64-setup.exe or the MSI | The installer is unsigned. If SmartScreen warns, select More info → Run anyway. If Windows asks to install or update WSL, allow it. |
| Debian or Ubuntu x86-64 | ai-security-scanner_0.4.0_amd64.deb | Install with sudo apt install ./ai-security-scanner_0.4.0_amd64.deb. |

The app prepares its own scanning runtime the first time a scan needs it; Docker is not required. Checksums and the tested limits of this release are on the v0.4.0 release page.

Open the app and choose one path:

Follow New scan → Review and start → Scan progress → Results. From Results, select Save or share report to open Share results, then Save HTML report. See the step-by-step path to an HTML report, including how to reopen a scan from My scans.

Use with Claude Code or Codex

Scans run in the desktop app on your own computer, and Start is a button in the app window. An agent on the same computer can install the app from the table abo that the computer is ready to scan, guide you through the app, explain the results, and save the HTML report. It does not need to build this repository. Both agents use the same skill:

| Agent | Repository skill |
| --- | --- |
| Claude Code | ai-security-scanner |
| Codex | ai-security-scanner |

Open this checkout in either agent and ask:

> Use the ai-security-scanner skill to install the app on this computer, check that it can scan, guide me through a scan, and save the final HTML report.

When the agent's sandbox blocks a download or the app, it asks you to allow that one command. A cloud agent cannot install anything on your computer, so it gives you the installer link instead. You choose the scope; the product selects applicable upstream checks. See what the agent needs.

One report for every selected asset

The report leads with:

Completed results remain a when an independent check fails. Reports can be reopened, compared with later runs, and exported as readable HTML or structured data.

Read a standard report

v0.4.0 · 2026-10-04: simulated assessment data in the real supported formats of all 25 adapters, processed through the production report pipeline. See 62 original findings and 42 in obser, with scanner sources and purposes. In is not counted as .

Both disclosure levels use the same simulated facts. Full disclosure keeps fictional asset and account context; secret stay masked in both. The standard product export is preserved, with a sample label and source appendix added.

What is new in v0.4.0

See the v0.4.0 release record.

Earlier changes in v0.3.1

See the release record for exact installer and the pending owner-run Microsoft 365 rerun.

Integrated tools

The v0.4.0 release integrates 25 upstream projects, including optional ZAP passi checks, Agentic Radar offline workflow in and bounded Garak model checks. Garak requires an exact HTTPS endpoint/model, fresh local key and explicit approval of its 54 fixed nati and provider charges. Research-only candidates are not current scan capabilities. See De status. The product runs only the tools that apply to each selected asset, keeps their original identifiers, se, e, and remediation, then organizes every completed result in the same report.

Selected assets and authorization → thin adapters → upstream scanners → output con → one standardized report organized by asset

Detection rules remain upstream. Product-owned prioritization, deduplication, and plain-language guidance live in the shared report layer.

Repositories, dependencies, and infrastructure as code

| Tool | What ai-security-scanner uses it for |
| --- | --- |
| Semgrep | Static analysis for risky code patterns using 1,493 pinned legacy upstream rules plus four product rules. |
| Gitleaks | Offline secret-pattern scanning with secret redacted from normal e. |
| TruffleHog | Offline filesystem secret detection; network is disabled. |
| Trivy | packages in recognized repository manifests and single-image OCI layouts using pinned offline data. |
| Grype | packages in repository snapshots and single-image OCI layouts using pinned offline data. |
| Checkov | Applicable infrastructure and configuration checks across the selected read-only snapshot. |
| KICS | Infrastructure-as-code misconfiguration checks from the upstream query pack. |
| Syft | Software component in and preserved SBOM output; in is not a result. |

The released catalog pins the Grype image to 0.117.0-4, digest sha256:56b0d675…. It can scan local repository dependencies for known ; a controlled repository scan recorded 81 Grype findings. That is a fixture result, not an expected count for every project. See the exact pin and recorded result.

Websites and internal systems

| Tool | What ai-security-scanner uses it for |
| --- | --- |
| Nuclei | Technology-aware, bounded read-only HTTP security checks from a pinned Nuclei Templates snapshot. |
| ZAP | Optional page-crawling and upstream passi checks for one appro. |
| Greenbone Open | Ser remote checks from a pinned Community Feed for exact appro and ports. |
| Naabu | Selected TCP-port reachability and exposure discovery; an open port is not a finding. |
| httpx | Bounded HTTP reachability and status metadata; it is not a scanner. |

Cloud and Microsoft 365

| Tool | What ai-security-scanner uses it for |
| --- | --- |
| Prowler | Narrow, exact-asset IAM configuration profiles for AWS, Azure, and GCP. |
| ScoutSuite | A reduced AWS IAM assessment rather than full ScoutSuite co. |
| Cloudsplaining | Excessi analysis o AWS IAM e. |
| CloudQuery | A fixed AWS IAM in; returned rows remain in rather than security findings. |
| Steampipe | AWS IAM user in; in fields do not become findings. |
| ScubaGear | A fixed Microsoft 365 security-baseline configuration profile. |
| Maester | A fixed Microsoft 365 security-configuration test profile. |

Kubernetes

| Tool | What ai-security-scanner uses it for |
| --- | --- |
| Kubescape | Offline configuration checks o selected local Kubernetes manifests. |
| kube-bench | CIS checks over an immutable node-configuration snapshot, without a pri li mount. |

AI workflows and MCP configuration

| Tool | What ai-security-scanner uses it for |
| --- | --- |
| Agentic Radar | Optional offline workflow in for LangGraph, CrewAI, n8n, OpenAI Agents, and AutoGen in one appro snapshot. It does not execute the workflow or contact a model. |
| Garak | Optional model behavior checks against one appro chat API/model with 54 native DAN/ANSI prompts, one-shot local keys and explicit inference limits/provider charges. |
| MCP Armor | Static checks over one approved MCP configuration snapshot for hardcoded credentials and excessive tool permissions, without starting or contacting an MCP server or loading a model. |

Discovery, in, SBOM generation, and the localhost TCP utility remain clearly separated from findings. The complete pinned , licenses, profiles, and execution boundaries are recorded in the engine catalog.

Exact scan boundaries and profile behavior are documented in Scanning scope.

Data and authorization

Cases, findings, and e stay on the device until an external source is connected or a report is exported. Local folders are copied into bounded read-only snapshots. Network scanners contact only the targets and ports confirmed on Review.

Use network scanning only for assets you own or are authorized to assess. The app does not apply remediation automatically.

Documentation

De

De requires Node.js 24 or newer, Rust 1.98, and the platform dependencies required by Tauri.

npm run dev opens a browser preview with sample data. Desktop scanning runs through the Tauri application.

npm run upstream:refresh -- --engine <id> produces an offline adapter refresh proposal bundle, and npm run upstream:propose -- --bundle <path> re- it. mechanical is the default deterministic path; the optional AI path is selected explicitly with --provider cli --ai-cli <executable>. --open-pr or --no-open-pr records whether the change should be sent back to this repository. These commands print commands for a human to run and ne them. The full procedure is in Engine maintenance.

License

Project-owned source is licensed under Apache-2.0. Third-party engines and data retain their own licenses; see THIRD_PARTY.md.

项目信息

开发语言：Rust

Star 数：10

技术要点

该项目属于安全软件方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Desktop app and Agent Skills that run established security scanners on your code, websites, and systems. One report.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。