personal-sentinel - Local, read-only Windows telemetry system for defe

项目说明

Sentinel

Sentinel V0.1 is a local, read-only Windows system obser as the first
layer of a personal defensi platform. It collects real system,
process, and network telemetry. It does not label activity as malicious,
block connections, terminate processes, or send data outside the computer.

Why V0.1 exists

Security detection is only as trustworthy as its telemetry. This keeps
the problem small enough to understand: collect operating-system facts, preserve
partial results when permissions or short-li cause errors, and show
the result through a testable CLI.

The collectors do not call one another. Models do not know about presentation.
The CLI composes those pieces. This is enough separation to support later event
normalization without in a message bus, database, API, or service layer.

Feature decisions

| Feature | Problem solved | Concepts learned | System and security value | Career value | Needed now? | How it is tested |
|---|---|---|---|---|---|---|
| Typed collection result | OS reads can partially fail | dataclasses, generics, contracts, UTC | prevents one inaccessible item from hiding all telemetry | backend and data contracts | Yes | success and partial-error unit tests |
| System collector | no concise host health view | CPU, RAM, disk, host identity | establishes system context and collector health | obser, DevOps, endpoint engineering | Yes | mocked metrics plus live status run |
| Process collector | no into executing programs | PID, PPID, process tree, CPU sampling, memory | raw endpoint telemetry for later baselines | SOC, EDR, platform engineering | Yes | parent correlation and access-denied tests |
| Network collector | no process-to-socket view | TCP/UDP, local/remote endpoint, state, ports | raw context for later network detections | networking, SOC, cloud security | Yes | local connection and permission-failure tests |
| CLI | telemetry needs a usable interface | argument parsing, formatting, exit codes | manual inspection without a web attack surface | backend and operations tooling | Yes | CLI output tests and manual runs |
| Structured logging | silent failures make results misleading | levels, JSON, rotation, privacy | records application health without command lines or secrets | SRE, De, incident diagnosis | Yes | tests exercise log creation; manual inspection |
| Database/API/dashboard/detection | no V0.1 problem requires them | later milestones | would add complexity before stable telemetry | later | No | deferred until acceptance criteria exist |

Prerequisites

Some network ownership details can require an elevated terminal. Sentinel treats
missing permissions as a warning; ele is not required for normal learning.
Do not run as Administrator merely to remove a harmless warning.

Setup (PowerShell)

If py is una, install Python 3.12+ from python.org and enable the
Python launcher during setup. The current repository was with the
official Python 3.14.7 distribution.

Usage

Application-health logs are written as JSON lines to logs/sentinel.log in the
current directory. Set SENTINEL_LOG_DIR to choose a different local directory.
Process command lines are held in the in-memory model but are not printed or
logged by V0.1 because arguments can contain credentials.

How to read the data

Tests

The tests replace operating-system calls with controlled examples. This makes
success, process exit, permission failure, parent correlation, and CLI behavior
repeatable without creating suspicious activity.

Manual learning lab

1. Run sentinel status twice. Explain why CPU changes while host identity does not.
2. Run sentinel processes --limit 30. Find Sentinel's Python process and identify its PPID.
3. Open Notepad, run the process command again, and explain the new PID without calling it a threat.
4. Start a local-only server with python -m http.server 8765 --bind 127.0.0.1 in a second terminal.
5. Run sentinel network, find port 8765, and identify LOCAL, REMOTE, LISTEN, and the owner PID.
6. Stop the local server with Ctrl+C, rerun the command, and explain the change.

You understand the milestone when you can predict these obser, explain
why a short-li may disappear during collection, and describe why a
single unusual process or port is e to in—not a malware .

Security and privacy boundaries

Current limitations

Definition of done for V0.1

Next milestone (not implemented)

V0.2 will normalize selected telemetry into security events and store them in
SQLite. It should begin only after V0.1 behavior and concepts can be explained
without relying on the code as a black box.

项目信息

开发语言：Python

Star 数：1

技术要点

该项目属于防护思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Local, read-only Windows telemetry system for defensive cybersecurity learning, monitoring and security engineering with Python.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。