Skip to content
View Whxuan0701's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Whxuan0701

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Whxuan0701/README.md

Wei Haoxuan

Harbin University of Science and Technology
Agent开发 / 后端开发
GitHub: @Whxuan0701

关注 AI Agent 工程、后端系统与可靠性,希望把 Agent 从“可以运行”推进到“可验证、可恢复、可观测”。

在研 APGFR: Prior-Guided Sparse-View Gaussian Reconstruction (3D Gaussian)

About Me

  • 学校:哈尔滨理工大学
  • 专业:大数据
  • 研究方向:Agent工程 / 大模型 / 计算机视觉 / 多模态Agent
  • 当前关注:AI Agent、skill、Agent Runtime、RAG
  • 技术栈:Python / TypeScript / Go / Java
  • 邮箱:641627652@qq.com

Open-source Snapshot

截至 2026-08-29,GitHub 公开记录中:

  • 33 个仓库提出 92 个 Pull Requests,其中 91 个是外部仓库贡献
  • 35 个已合并48 个仍开放9 个已关闭;开放项中包含仍处于 Draft 的工作
  • 6 个 Apache 仓库提出 9 个 PR:3 个已合并、4 个仍开放、2 个已关闭

Representative Contributions

1. Agent Runtime: bounded execution and recovery

buildwithclaude GitHub Stars 中实现并合并:

  • checkpointed-agent-loop:为长任务提供版本化检查点、有限状态机、重试预算、原子写入、恢复入口和验证证据;脚本只负责状态,不执行命令或访问网络。
  • rag-evaluation-harness:实现确定性、离线的 RAG 评测工具,计算 Recall@K、MRR、Context Precision、Citation Coverage/Validity,并用退出码支持 CI 阈值门禁。

在 Apache Agent 框架 Burr GitHub Stars 中提交的进行中贡献:

  • InMemoryExecutionRecorder:通过生命周期钩子记录 Agent 的 action、输入、结果、异常和状态变化,用于本地回放与测试。
  • Graph.to_dict():提供无需 Graphviz 或 tracking 依赖的确定性图结构导出。
  • max_steps execution budget:为同步/异步执行与迭代接口增加单次执行步数预算,在 Agent Loop 未自然结束时安全失败并保留可恢复状态。

2. Multi-agent isolation and orchestration safety

kagent GitHub Stars 中实现并合并:

llama-agents GitHub Stars 中实现并合并:

3. MCP client and inspection tooling

mcp-use GitHub Stars 中实现并合并:

其他 MCP/SDK 方向的贡献还覆盖工具命名空间、任务轮询、Roots Handler、MCP 初始化就绪检查、机器可读输出与跨 Agent Trace 传播,完整状态见 PR Ledger

4. Agent evaluation, evidence and failure testing

agentic-awesome-skills GitHub Stars 中合并 5 项 Agent 工程方法:

这些工作把多 Agent 编排视为分布式状态机,重点检查所有权、租约与 fencing、版本化状态、分支 Join、重试与幂等、取消、预算、检查点恢复以及“验证制品与部署制品是否一致”。

GenAI_Agents GitHub Starsagents-towards-production GitHub Stars 中的 6 个 PR 也已合并,覆盖 Human-in-the-Loop 审批 Agent 教程、Trace-based 评测教程、Notebook 贡献校验,以及 Agent 教程依赖声明和 FastAPI 流式响应测试修复。

5. RAG and code intelligence reliability

code-graph-rag GitHub Stars 中合并 3 个可靠性修复:

这些改动避免文件系统顺序、陈旧图位置和非法范围影响 Agent 的检索结果,使下游 RAG/代码理解流程获得更稳定、可信的上下文。

6. Search, CLI and workflow engineering

  • buildwithclaude #270 GitHub Stars#271#272:处理 YAML 日期归一化、搜索对象 ID 冲突,以及 Plugin/Skill 同名时的检索结果身份保持。
  • OpenCLI #2298 GitHub Stars#2299#2300:增强无效 Manifest 回退、<1.0.0 caret 版本范围及整数参数校验。
  • planning-with-files #222 GitHub Stars#223#224:修复 Copilot PowerShell Hook、禁用开关传播与 Python 解释器一致性。
  • Needle #93 GitHub Stars:同步 finetune CLI 的模型、并发、checkpoint 目录和 adapter 路径文档,并与源码参数保持一致。

7. Apache contributions

已合并:

审阅中:

Engineering Principles

  • Bounded execution:循环、重试和并发必须有明确预算及终止条件。
  • Recoverable state:检查点要可验证、可恢复,并区分业务状态与尝试记录。
  • Evidence before claims:完成状态必须绑定测试、Trace、Artifact 或可复现输出。
  • Protocol compatibility:扩展 MCP/SDK 接口时保留兼容路径,明确取消、超时与错误语义。
  • Deterministic behavior:排序、标识符、序列化和状态转换应可重复验证。
  • Honest boundaries:Open、Draft、Closed 与 Merged 分开呈现;局部测试不等同于全仓库或生产验证。

Current Focus

  • 3D Gaussian
  • ReAct
  • Dynamic Patch Perception

Contact

完整的 91 个 PR、仓库分布及当前状态:OPEN_SOURCE_PR_LEDGER.md

Pinned Loading

  1. langgenius/dify langgenius/dify Public

    Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without …

    TypeScript 154k 24.4k

  2. OpenHands/OpenHands OpenHands/OpenHands Public

    🙌 OpenHands: AI-Driven Development

    TypeScript 85.9k 11.3k

  3. microsoft/autogen microsoft/autogen Public

    A programming framework for agentic AI

    Python 60.8k 9.2k

  4. crewAIInc/crewAI crewAIInc/crewAI Public

    Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

    Python 58k 8.3k

  5. calesthio/OpenMontage calesthio/OpenMontage Public

    World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your AI coding assistant into a full video pro…

    Python 55.4k 6.9k

  6. run-llama/llama_index run-llama/llama_index Public

    LlamaIndex is the leading document agent and OCR platform

    Python 52k 8.1k