Agentic AI Researcher @ Moonshot AI (Kimi) | MS CS @ HKU | Champion, Shanghai Global AI Contest | 3x ACM-ICPC Silver Medalist | Former Intern @ Baidu, Maimai, Kuaishou
- 80+ merged upstream PRs, including fixes in vLLM, Hugging Face Transformers, Microsoft Agent Framework, AstrBot, Mooncake, Inspect AI, LiteLLM, and Triton.
- 18 public projects, led by CoreCoder, RepoWiki, FindJobs-Agent, ContractGuard, AgentProbe, GitSense, and MTSIR3-GAN.
| Area | Project | Stars | Notes |
|---|---|---|---|
| Coding agents / evals | CoreCoder | 984+ | 512K lines of Claude Code → 1,400 lines of Python. Every key architectural pattern, runnable. Any LLM. 7 architecture deep-dive articles. |
| Codebase maps | RepoWiki | 136+ | Open-source DeepWiki alternative: CLI/browser wiki generation, PageRank file ranking, Mermaid diagrams, reading guides. |
| Applied agents | FindJobs-Agent | 223+ | LLM-powered job toolkit: skill gap analysis, mock interviews, resume optimization. |
| Applied agents | ContractGuard | 127+ | AI agent that reviews contracts for red flags before you sign. |
| Coding agents / evals | AgentProbe | 38+ | Pytest plugin for regression-testing AI agents: snapshot baselines, semantic comparison, mock LLMs. |
| Coding agents / evals | CodeJoust | 6+ | CLI arena for AI coding agents: isolated git worktree runs, test/cost/diff/time scoring, pip install codejoust. |
| Coding agents / evals | LiteBench | 3+ | Pip-installable LLM/agent benchmark CLI + web dashboard, with agent mode, custom YAML, LLM-as-judge, and 100+ litellm models. |
| Coding agents / evals | AnyCoder | 21+ | AI coding agent CLI supporting 100+ LLMs via litellm. ~1,450 lines Python, dangerous command blocking, parallel tool execution, session persistence. |
| Codebase maps | GitSense | 68+ | AI-powered contribution finder and repo radar: find matching issues, then check PR merge signals. |
| Codebase maps | RuleForge | 76+ | Auto-generate AI assistant rules (CLAUDE.md, .cursorrules) from codebase analysis. |
| Codebase maps | CodeABC | 57+ | AI code reader for non-programmers: drag in a project, get plain-language explanations with hover annotations. |
| LLM tooling | MCPReady | 1+ | CI gate for MCP servers: handshake, tools/list, schema checks, secret-leak scan, reports, GitHub Action. |
| LLM tooling | TokenTracker | 46+ | Drop-in LLM cost tracker: change one import line and see where the money goes. |
| LLM tooling | PromptDiff | 25+ | Semantic diff for LLM prompts, compare prompt versions like git diff. |
| LLM tooling | BatchLLM | 26+ | Batch LLM processing: CSV in, processed CSV out, with retries and cost tracking. |
| Research / quant | MTSIR3-GAN | 135+ | Adapting R3GAN (NeurIPS 2024) to multivariate time series imputation + adversarial refinement study. |
| Research / quant | DRL-MultiFactorTrading | 75+ | Deep RL trading strategies with Double DQN and multi-factor alpha signals. |
| Research / quant | Arxiv Paper Coding Agent | 39+ | Multi-LLM crew that reads a paper and produces a working implementation. |
| Project | PR | What I Fixed |
|---|---|---|
| vLLM (81.2k★) | #40789 | Support tuple model outputs in the V1 ubatch wrapper so DBO and speculative decoding stop crashing on tuple-returning models |
| vLLM (81.2k★) | #37884 | RoBERTa position_ids in-place accumulation on CUDA graph padding: BGE-M3 crash after ~4000 requests |
| vLLM (81.2k★) | #37727 | Fix Responses API instructions leaking across turns via previous_response_id chain |
| vLLM (81.2k★) | #37699 | Fix weight offloading ignoring VLLM_WEIGHT_OFFLOADING_DISABLE_PIN_MEMORY in prefetch offloader |
| vLLM (81.2k★) | #37301 | Base64 JPEG video frames returning empty metadata: populate frame count, fps, duration |
| vLLM (81.2k★) | #38732 | Fix bench_serve UTF-8 decode crash on split multi-byte chars in streaming chunks |
| vLLM (81.2k★) | #40749 | Helped unblock async PP fix with independent repro evidence for the last-rank sampled-token receive assertion |
| Transformers (161k★) | #44710 | Fix AutoProcessor.from_pretrained silently dropping hub kwargs (revision, token, etc.) |
| Mooncake (5.4k★) | #1629 | GB200 MNNVL EP hang: cudaMalloc → cuMemCreate(FABRIC) + cuMemMap for cross-node NVLink |
| Mooncake (5.4k★) | #1831 | TENT NVLink IPC fix: use base pointer for sub-allocated GPU tensors, porting #1622 fix to TENT path |
| Mooncake (5.4k★) | #1644 | MNNVL warmup hang: skip redundant handshake for fabric-connected nodes |
| Mooncake (5.4k★) | #1728 | Hard pin for eviction-protected objects: model weights never get evicted, const field + BatchEvict skip + backward-compat serialization |
| Mooncake (5.4k★) | #1719 | Add ObjectDataType metadata classification for KV cache, weights, tensors, and snapshots, with backward-compatible serialization and Python bindings |
| Mooncake (5.4k★) | #1825 | Fix P2PClientService::Put silently swallowing write errors: propagate actual error codes for non-idempotent failures |
| Mooncake (5.4k★) | #1626 | Silenced error log spam for non-memory replicas in metadata store |
| LiteLLM (48.4k★) | #26401 | Fix LITELLM_LOG=INFO missing verbose_logger: proxy INFO logs now include all verbose logger sources |
| Triton (19.3k★) | #9613 | Fix AxisInfo correctness: signed constants, unknown shift divisibility, and shift UB guards |
| SGLang (28.3k★) | #20739 | Fix hybrid_linear_attn_backend crash when used with ngram speculative decoding |
| FlashInfer (5.7k★) | #2756 | Fix autotuner crash when input tensor is None: proper None-checking for optional inputs (fixes #2749) |
| FlashInfer (5.7k★) | #2772 | Fix compilation error: add missing <optional> header for std::optional usage in CUTLASS headers |
| LMCache (8.4k★) | #3282 | Support HND GPU KV formats in MP KV transfer so alternate KV tensor layouts are handled correctly |
| Project | PR | What I Fixed |
|---|---|---|
| OpenClaw (375k★) | #41259 | Propagate memory directory creation failures instead of continuing after a failed ensureDir |
| OpenClaw (375k★) | #41271 | Log auth profile resolution failures instead of swallowing silently |
| Microsoft Agent Framework (10.8k★) | #5462 | Fix background=True + tools infinite-retrieve loop: clear completed continuation state so tool results get posted |
| Microsoft Agent Framework (10.8k★) | #5861 | Preserve structured Foundry handoff arguments so delegated agent calls survive response conversion |
| Microsoft Agent Framework (10.8k★) | #5974 | Include Foundry agent tool definitions in eval mappings so evaluator runs can call code-defined tools |
| Microsoft Agent Framework (10.8k★) | #5815 | Forward MCP tool-call metadata from tools/list into call_tool instrumentation so traces keep tool annotations |
| Microsoft Agent Framework (10.8k★) | #5778 | Declare Magentic protocol messages so orchestrator chat/reset signals deserialize instead of failing on unknown types |
| Microsoft Agent Framework (10.8k★) | #5808 | Fix handoff message role mutation so retries reuse sanitized messages without leaking role changes |
| Microsoft Agent Framework (10.8k★) | #5800 | Avoid AG-UI tool result message id collisions when providers omit update ids |
| Microsoft Agent Framework (10.8k★) | #5799 | Add handoff workflow naming metadata so names and descriptions flow through builder and hosting registration |
| Microsoft Agent Framework (10.8k★) | #5976 | Fix sequential workflow sample output so all participant responses are shown from non-streaming results |
| AstrBot (33.2k★) | #6596 | Multimodal token counting: images, audio, chain-of-thought were invisible to context compression |
| AstrBot (33.2k★) | #8089 | Route image requests to a vision-capable fallback provider when the primary provider cannot accept image input |
| AstrBot (33.2k★) | #7758 | Apply empty-assistant message filter to streaming OpenAI path: strict providers no longer 400 on reasoning-only turns |
| AstrBot (33.2k★) | #8195 | Skip empty LLM summaries so context compression keeps the original history instead of inserting blank placeholders |
| AstrBot (33.2k★) | #8073 | Fix image-only KB retrieval: skip blank prompts so embedding APIs are not called with empty text |
| AstrBot (33.2k★) | #8119 | Pass image inputs through active replies so image-triggered mentions reach the LLM as image URLs |
| AstrBot (33.2k★) | #7003 | SSE heartbeat for WebChat: long context compression killed the connection |
| AstrBot (33.2k★) | #6581 | Context truncation dropping the only user message: causes 400 from Zhipu/Gemini |
| AstrBot (33.2k★) | #7216 | Fix Gemini tool call 400: wrap plain-text tool results as Protobuf Struct |
| AstrBot (33.2k★) | #7407 | Fix Gemini native search 400 when no function tools: skip FunctionCallingConfig |
| AstrBot (33.2k★) | #7196 | Fix Gemini thinking parts leaking into user-facing response |
| AstrBot (33.2k★) | #6710 | Skills-like re-query dropping image captions: extra_user_content_parts not forwarded |
| AstrBot (33.2k★) | #8255 | Support RST and AsciiDoc knowledge uploads by keeping backend parser checks and dashboard file validation in sync |
| AstrBot (33.2k★) | #6474 | SQLite database is locked under concurrent writes: added busy timeout |
| AstrBot (33.2k★) | #8175 | Surface WeChat OA media send failures instead of reporting success after prepare/send timeouts |
| AstrBot (33.2k★) | #8172 | Prefer bundled dashboard assets over stale data dist so releases do not serve an outdated WebUI |
| AstrBot (33.2k★) | #8061 | Keep Discord startup alive when command sync hits the daily create quota; only quota errors become warnings |
| AstrBot (33.2k★) | #7217 | Fix qwen3-rerank response parsing: handle both old and new Dashscope API formats |
| AstrBot (33.2k★) | #7537 | Prevent Telegram media group exceptions from being silently swallowed by APScheduler |
| AstrBot (33.2k★) | #7398 | Fix Telegram sendMessageDraft spamming 400 errors on empty text in streaming mode |
| AstrBot (33.2k★) | #6656 | /stop follow-up race: agent_stop flag not checked during follow-up capture |
| AstrBot (33.2k★) | #6551 | Fix empty content causing Grok 400: set content to None when empty |
| AstrBot (33.2k★) | #6313 | Null choices guard: OpenAI API returning None instead of empty list |
| AstrBot (33.2k★) | #6527 | Fix LLM tool selection: rewrote ambiguous Upload/Download descriptions |
| AG-UI (13.9k★) | #1735 | Avoid stale ADK session writes after human-in-the-loop tool calls |
| AG-UI (13.9k★) | #1769 | Make proto generation cross-platform: replace Unix-only mkdir -p with a Node script and Windows .CMD plugin shim |
| Pydantic AI (17.3k★) | #5474 | Accept providerExecuted and title on Vercel AI dynamic-tool message parts so strict validation preserves provider metadata |
| LiveKit Agents (10.7k★) | #5820 | Recreate Anthropic streaming requests on retry so transient stream creation failures do not re-await the same coroutine |
| Pipecat (12.5k★) | #4553 | Serialize interruption frames through protobuf transports so realtime voice-agent interruptions survive transport hops |
| Inspect AI (2.1k★) | #3982 | Preserve wrapped OpenAI reasoning payloads in the agent bridge so encrypted provider-native reasoning items survive transcript conversion |
| Inspect AI (2.1k★) | #3975 | Preserve call-site generation defaults when role model overrides are resolved, so eval roles can switch models without dropping config like max tokens or reasoning effort |
| Inspect AI (2.1k★) | #3896 | Fix filestore recovery append mode: preserve carried message/tool-call pools without rehashing old segments |
| Inspect AI (2.1k★) | #3941 | Route Bedrock Nova top_k through the inference config instead of dropping the sampling control |
| Inspect AI (2.1k★) | #3902 | Parse OpenRouter reasoning_details in OpenAI-compatible responses instead of surfacing Python repr blocks |
| Inspect AI (2.1k★) | #3924 | Reject unknown GenerateConfig fields early so misspelled generation options do not get silently ignored |
| Inspect AI (2.1k★) | #3895 | Honor COLUMNS for TERM=dumb: log output no longer hard-wraps at Rich's default 80 columns |
| MCP Go SDK (4.6k★) | #962 | Reject duplicate initialize requests so MCP sessions keep protocol state consistent after initialization |
| OpenHands SDK (754★) | #3248 | Serialize LiteLLM modify_params updates with an RLock so concurrent completions do not leak global parameter state |
| OpenHands SDK (754★) | #3247 | Validate git workspaces with git rev-parse --git-dir so broken nested repos do not crash /api/git/changes |
| OpenHands SDK (754★) | #3225 | Write remote completion logs as UTF-8 so non-ASCII output survives local replay and debugging |
| ClawTeam (5.3k★) | #1 | First PR: 122 tests, CI, team templates, config bugfixes, task duration tracking |
| ClawTeam (5.3k★) | #40 | Pluggable TaskStore: extract task persistence into swappable backend abstraction |
| ClawTeam (5.3k★) | #32 | Gemini CLI support: spawn, permissions, prompt injection for both backends |
| ClawTeam (5.3k★) | #36 | Kimi CLI support: spawn backend, permission handling, 3 new test cases |
| OpenHarness (13.2k★) | #185 | TUI tab-completion: fix cursor jump, strip trailing space, accept /quit alias |
| Project | PR | What I Fixed |
|---|---|---|
| Phoenix (9.9k★) | #13245 | Keep the generative model fetch cursor monotonic so lower-id updates cannot make later polling skip newer model changes |
| Phoenix (9.9k★) | #13210 | Return NotFound-style errors for invalid GraphQL node ids instead of leaking decoder failures to clients |
| Phoenix (9.9k★) | #13139 | Surface playground validation errors instead of returning empty subscription payloads |
| Phoenix (9.9k★) | #13261 | Refresh span annotation notes after create so the UI shows newly added notes without a manual reload |
| LightRAG (35.9k★) | #3031 | Extract Docling async markdown results from the response envelope so RAG chunks contain clean document text |
| LightRAG (35.9k★) | #2796 | Fix None file_path propagating as unknown_source: fill gaps left by #2793 |
| LightRAG (35.9k★) | #3123 | Sync API documentation colors with the dark theme so endpoint examples stay readable in dark mode |
| Project | PR | What I Fixed |
|---|---|---|
| Microsoft Recommenders (21.7k★) | #2322 | Honor benchmark recommendation-count args so evaluation scripts generate the requested top-k results |
LinkedIn: https://www.linkedin.com/in/yufenghe
Moonshot AI (Kimi) AI Agent 研究员 | 港大计算机硕士 | 上海全球AI大赛冠军 | 三次获ACM-ICPC银牌 | 曾在百度、脉脉、快手的AI 研发岗实习
- 80+ 个上游 PR 已 merged，包括 vLLM、Hugging Face Transformers、Microsoft Agent Framework、AstrBot、Mooncake、Inspect AI、LiteLLM、Triton 等项目里的生产级 bug / 行为修复。
- 18 个公开项目，代表项目包括 CoreCoder、RepoWiki、FindJobs-Agent、ContractGuard、AgentProbe、GitSense、MTSIR3-GAN。
| 方向 | 项目 | Stars | 简介 |
|---|---|---|---|
| Coding agents / 评测 | CoreCoder | 984+ | Claude Code 51万行源码 → 1400行 Python 核心重写，支持任意大模型，附 7 篇架构导读。 |
| 代码库地图 | RepoWiki | 136+ | 开源 DeepWiki 替代品：CLI 或浏览器生成仓库 wiki，PageRank 文件排名、Mermaid 架构图、阅读指南。 |
| 应用型 Agent | FindJobs-Agent | 223+ | LLM 求职工具箱：技能差距分析、模拟面试、简历优化。 |
| 应用型 Agent | ContractGuard | 127+ | AI 合同审查 Agent，签字前帮你找出霸王条款。 |
| Coding agents / 评测 | AgentProbe | 38+ | AI Agent 回归测试 pytest 插件：快照基线、语义比较、Mock LLM。 |
| Coding agents / 评测 | CodeJoust | 6+ | AI 编程 Agent 擂台：独立 git worktree 运行，按测试通过率、成本、diff 大小、耗时打分。 |
| Coding agents / 评测 | LiteBench | 3+ | LLM / Agent benchmark 轻量 CLI + Web 面板：agent 模式、自定义 YAML、LLM judge、100+ litellm 模型。 |
| Coding agents / 评测 | AnyCoder | 21+ | 终端 AI 编程 Agent，通过 litellm 支持 100+ 大模型，~1450 行 Python，危险命令拦截、并行执行、会话持久化。 |
| 代码库地图 | GitSense | 68+ | AI 开源贡献发现器 + 仓库雷达：找匹配 issue，也评估 PR 合入友好度。 |
| 代码库地图 | RuleForge | 76+ | 从代码库自动生成 AI 助手规则（CLAUDE.md、.cursorrules）。 |
| 代码库地图 | CodeABC | 57+ | 面向非程序员的 AI 代码阅读器：拖入项目，生成大白话解释和悬浮批注。 |
| LLM 工具链 | MCPReady | 1+ | MCP Server CI 质量门禁：握手、tools/list、schema 检查、明显 token 泄露扫描、报告和 GitHub Action。 |
| LLM 工具链 | TokenTracker | 46+ | LLM 开销追踪，改一行 import 就能看到钱花在哪里。 |
| LLM 工具链 | PromptDiff | 25+ | LLM prompt 语义 diff 工具，像 git diff 一样对比 prompt 版本。 |
| LLM 工具链 | BatchLLM | 26+ | LLM 批量处理：CSV 进 CSV 出，自动重试 + 费用追踪。 |
| 研究 / 量化 | MTSIR3-GAN | 135+ | 将 R3GAN (NeurIPS 2024) 适配到多变量时序插补 + 对抗精炼实证研究。 |
| 研究 / 量化 | DRL-MultiFactorTrading | 75+ | 深度强化学习量化交易，Double DQN + 多因子 Alpha。 |
| 研究 / 量化 | Arxiv Paper Coding Agent | 39+ | 多 LLM 协作 Agent，读论文自动生成可运行代码。 |
| 项目 | PR | 修了啥 |
|---|---|---|
| vLLM (81.2k★) | #40789 | 修复 V1 ubatch wrapper 不支持 tuple model outputs：DBO / speculative decoding 不再因 tuple 返回值崩溃 |
| vLLM (81.2k★) | #37884 | 修复 RoBERTa position_ids 原地累积溢出：BGE-M3 约 4000 请求后 CUDA graph padding crash |
| vLLM (81.2k★) | #37727 | 修复 Responses API 的 instructions 通过 previous_response_id 链泄漏到后续轮次 |
| vLLM (81.2k★) | #37699 | 修复 weight offloading 忽略 VLLM_WEIGHT_OFFLOADING_DISABLE_PIN_MEMORY 环境变量 |
| vLLM (81.2k★) | #37301 | 修复 base64 JPEG 视频帧返回空 metadata：补充帧数、fps、时长 |
| vLLM (81.2k★) | #38732 | 修复 bench_serve 流式响应拆分多字节 UTF-8 字符导致 decode 崩溃 |
| vLLM (81.2k★) | #40749 | 补充独立复现证据，帮助确认 async PP last-rank sampled-token receive assertion 的真实可达路径 |
| Transformers (161k★) | #44710 | 修复 AutoProcessor.from_pretrained 静默丢弃 hub kwargs（revision、token 等） |
| Mooncake (5.4k★) | #1629 | GB200 MNNVL EP hang：cudaMalloc → cuMemCreate(FABRIC) + cuMemMap 跨节点 NVLink 通信 |
| Mooncake (5.4k★) | #1831 | TENT NVLink IPC 修复：sub-allocated GPU tensor 使用 base pointer，将 #1622 修复移植到 TENT 路径 |
| Mooncake (5.4k★) | #1644 | MNNVL warmup hang：跳过 fabric 连接节点的冗余握手 |
| Mooncake (5.4k★) | #1728 | Hard pin 驱逐保护：模型权重永不被驱逐，const 字段 + BatchEvict 跳过 + 向后兼容序列化 |
| Mooncake (5.4k★) | #1719 | 新增 ObjectDataType 元数据分类：KV cache、weights、tensors 等对象类型可在 metadata、snapshot 和 Python binding 中稳定传递 |
| Mooncake (5.4k★) | #1825 | 修复 P2PClientService::Put 静默吞掉写入错误：传播实际错误码给调用方 |
| Mooncake (5.4k★) | #1626 | 修复非内存副本的错误日志刷屏 |
| LiteLLM (48.4k★) | #26401 | 修复 LITELLM_LOG=INFO 漏设 verbose_logger：proxy INFO 日志不再静默丢失 |
| Triton (19.3k★) | #9613 | 修复 AxisInfo 正确性：有符号常量、未知 shift 和 shift UB 都保守处理 |
| SGLang (28.3k★) | #20739 | 修复 hybrid_linear_attn_backend 与 ngram 投机采样同时使用时崩溃 |
| FlashInfer (5.7k★) | #2756 | 修复 autotuner 在输入 tensor 为 None 时崩溃（fixes #2749） |
| FlashInfer (5.7k★) | #2772 | 修复编译错误：CUTLASS 头文件缺少 <optional> include 导致 std::optional 未定义 |
| LMCache (8.4k★) | #3282 | 支持 MP KV transfer 中的 HND GPU KV 格式，让不同 KV tensor layout 能被正确处理 |
| 项目 | PR | 修了啥 |
|---|---|---|
| OpenClaw (375k★) | #41259 | 目录创建失败时向上传递错误，避免 memory ensureDir 静默失败后继续执行 |
| OpenClaw (375k★) | #41271 | 认证配置解析失败时记录日志而非静默吞掉 |
| Microsoft Agent Framework (10.8k★) | #5462 | 修复 background=True + tools 无限 retrieve loop：清掉已完成的 continuation 状态，让 tool results 正常提交 |
| Microsoft Agent Framework (10.8k★) | #5861 | 修复 Foundry handoff 参数序列化，delegated agent 调用不再丢失结构化参数 |
| Microsoft Agent Framework (10.8k★) | #5974 | 让 Foundry agent eval 映射保留工具定义，评测运行能正常调用代码定义的 tools |
| Microsoft Agent Framework (10.8k★) | #5815 | 转发 MCP tools/list 元数据到 call_tool instrumentation，让 trace 保留工具注解 |
| Microsoft Agent Framework (10.8k★) | #5778 | 补齐 Magentic protocol 的 chat/reset message 声明，避免 orchestrator 反序列化未知类型失败 |
| Microsoft Agent Framework (10.8k★) | #5808 | 修复 handoff message role 原地修改：重试复用消息时不再把角色变更泄漏到后续流程 |
| Microsoft Agent Framework (10.8k★) | #5800 | 修复 AG-UI tool result message id 冲突：provider 省略 update id 时仍能生成独立 fallback id |
| Microsoft Agent Framework (10.8k★) | #5799 | 补齐 Handoff workflow 的 name/description 元数据传递，让 builder 和 hosting 注册都能命名工作流 |
| Microsoft Agent Framework (10.8k★) | #5976 | 修复 sequential workflow 示例输出，让非 streaming 结果展示所有参与 agent 的回复 |
| AstrBot (33.2k★) | #6596 | 多模态 token 计数：图片/音频/思考链对 context 压缩不可见的问题 |
| AstrBot (33.2k★) | #8089 | 图片请求在主 provider 不支持视觉输入时自动切到 vision fallback provider |
| AstrBot (33.2k★) | #7758 | 修复 OpenAI streaming 路径复用 empty-assistant 过滤：reasoning-only 历史不再让严格 provider 返回 400 |
| AstrBot (33.2k★) | #8195 | 修复空 LLM summary：压缩器在模型返回空摘要时保留原始历史，不再插入空 summary 占位 |
| AstrBot (33.2k★) | #8073 | 修复纯图片/表情消息的知识库空 prompt：空白文本时跳过 KB 检索，避免 embedding API 400 |
| AstrBot (33.2k★) | #8119 | 修复 active reply 图片输入丢失：把图片组件转成 image URLs 传给 LLM，保留原有文本 prompt 行为 |
| AstrBot (33.2k★) | #7003 | 修复 WebChat 长响应断连：SSE 心跳保活，context 压缩期间不再超时 |
| AstrBot (33.2k★) | #6581 | 修复截断器丢失唯一 user 消息导致智谱/Gemini 返回 400 |
| AstrBot (33.2k★) | #7216 | 修复 Gemini tool call 400：纯文本 tool result 包装为 Protobuf Struct |
| AstrBot (33.2k★) | #7407 | 修复 Gemini 原生搜索无 function tools 时 400：跳过 FunctionCallingConfig |
| AstrBot (33.2k★) | #7196 | 修复 Gemini thinking parts 泄漏到用户可见的消息内容中 |
| AstrBot (33.2k★) | #6710 | 修复 skills-like re-query 丢失图片描述：extra_user_content_parts 未传递 |
| AstrBot (33.2k★) | #8255 | 支持 RST / AsciiDoc 知识库上传：后端解析白名单和前端文件校验同步扩展，避免支持的文档格式被入口拦截 |
| AstrBot (33.2k★) | #6474 | 修复 SQLite 并发写入 database is locked，添加 busy timeout |
| AstrBot (33.2k★) | #8175 | 修复 WeChat OA 媒体发送失败仍上报成功：prepare/send 超时或失败时向调用方抛出明确错误 |
| AstrBot (33.2k★) | #8172 | 修复 release 使用过期 dashboard dist 的问题：优先加载 bundled WebUI 资产，避免发版后仍展示旧前端 |
| AstrBot (33.2k★) | #8061 | 修复 Discord 启动同步命令触发 daily create quota 时打断 bot 启动：只把配额错误降级为 warning，其它异常继续抛出 |
| AstrBot (33.2k★) | #7217 | 修复 qwen3-rerank 响应解析：兼容新旧百炼 API 格式 |
| AstrBot (33.2k★) | #7537 | 修复 Telegram media group 异常被 APScheduler 静默吞掉：try/except + EVENT_JOB_ERROR listener |
| AstrBot (33.2k★) | #7398 | 修复 Telegram 流式模式下空文本 sendMessageDraft 导致 400 错误刷屏 |
| AstrBot (33.2k★) | #6656 | 修复 /stop 后新消息仍被 follow-up 捕获的竞态条件 |
| AstrBot (33.2k★) | #6551 | 修复空 content 导致 Grok 400：content 为空时设为 None |
| AstrBot (33.2k★) | #6313 | 修复 OpenAI API 返回 None choices 导致崩溃 |
| AstrBot (33.2k★) | #6527 | 修复 LLM 工具选择：重写模糊的 Upload/Download 描述 |
| AG-UI (13.9k★) | #1735 | 修复 HITL tool call 后 ADK session 可能被旧状态回写的问题 |
| AG-UI (13.9k★) | #1769 | 将 proto generation 从 Unix-only mkdir -p 改为跨平台 Node 脚本，兼容 Windows .CMD plugin shim |
| Pydantic AI (17.3k★) | #5474 | 兼容 Vercel AI dynamic-tool 消息里的 providerExecuted 和 title 字段，保留 provider 元数据而不是被严格校验拒绝 |
| LiveKit Agents (10.7k★) | #5820 | 修复 Anthropic streaming retry：瞬时建流失败后重新创建 stream，不再重复 await 同一个 coroutine |
| Pipecat (12.5k★) | #4553 | 修复 protobuf transport 丢失 interruption frame：实时语音 agent 的打断事件跨传输后仍能保留 |
| Inspect AI (2.1k★) | #3982 | 保留 agent bridge 转换中的 OpenAI wrapped reasoning payload，避免 encrypted provider-native reasoning item 在 transcript 转换时丢失 |
| Inspect AI (2.1k★) | #3975 | 修复 role model override 合并配置：角色切换模型时保留调用侧 GenerateConfig 默认值，不再丢掉 max tokens / reasoning effort 等设置 |
| Inspect AI (2.1k★) | #3896 | 修复 filestore recovery append 模式：保留已恢复 message/tool-call 池，同时避免重复哈希旧片段 |
| Inspect AI (2.1k★) | #3941 | 修复 Bedrock Nova top_k 路由：把采样控制传到 inference config，而不是被静默丢弃 |
| Inspect AI (2.1k★) | #3902 | 修复 OpenAI-compatible 响应里的 OpenRouter reasoning_details：解析为可读 reasoning 文本，而不是暴露 Python repr |
| Inspect AI (2.1k★) | #3924 | 让未知 GenerateConfig 字段提前报错，避免拼错的生成参数被静默忽略 |
| Inspect AI (2.1k★) | #3895 | 修复 TERM=dumb 下终端宽度：尊重 COLUMNS，日志输出不再固定按 Rich 默认 80 列硬换行 |
| MCP Go SDK (4.6k★) | #962 | 拒绝重复 initialize 请求，让 MCP session 初始化后的协议状态保持一致 |
| OpenHands SDK (754★) | #3248 | 用 RLock 串行化 LiteLLM modify_params 的保存、设置和恢复，避免并发 completion 泄漏全局参数状态 |
| OpenHands SDK (754★) | #3247 | 用 git rev-parse --git-dir 验证 git workspace，避免坏的 nested repo 打崩 /api/git/changes |
| OpenHands SDK (754★) | #3225 | 按 UTF-8 写入 remote completion 日志，避免中文和非 ASCII 输出在本地回放、排查时损坏 |
| ClawTeam (5.3k★) | #1 | 首个 PR：122 个测试、CI、团队模板、config 修复、任务耗时追踪 |
| ClawTeam (5.3k★) | #40 | 可插拔 TaskStore：将任务持久化抽取为可替换的后端抽象层 |
| ClawTeam (5.3k★) | #32 | Gemini CLI 支持：spawn、权限、prompt 注入双 backend 适配 |
| ClawTeam (5.3k★) | #36 | Kimi CLI 支持：spawn backend、权限处理、3 个新测试 |
| OpenHarness (13.2k★) | #185 | TUI tab 补全三合一修复：光标跳回、去除尾部空格、接受 /quit 别名 |
| 项目 | PR | 修了啥 |
|---|---|---|
| Phoenix (9.9k★) | #13245 | 修复 generative model fetch 游标回退：低 id 更新不会让后续轮询跳过更新的模型变更 |
| Phoenix (9.9k★) | #13210 | 修复非法 GraphQL node id 的错误处理：返回 NotFound 风格错误，而不是把 decoder failure 泄漏给客户端 |
| Phoenix (9.9k★) | #13139 | 修复 Playground 校验错误被吞掉的问题：失败时返回明确错误，而不是空 subscription payload |
| Phoenix (9.9k★) | #13261 | 修复 span annotation note 新建后的刷新问题：创建成功后立即重新拉取列表，前端不再需要手动刷新才能看到新 note |
| LightRAG (35.9k★) | #3031 | 从 Docling 异步结果 envelope 中抽取 Markdown 正文，避免 RAG chunk 混入 JSON/base64 噪声 |
| LightRAG (35.9k★) | #2796 | 修复 None file_path 传播为 unknown_source：补 #2793 遗漏的处理层 |
| LightRAG (35.9k★) | #3123 | 同步 API 文档暗色主题配色，避免 endpoint 示例在 dark mode 下变得不可读 |
| 项目 | PR | 修了啥 |
|---|---|---|
| Microsoft Recommenders (21.7k★) | #2322 | 修复 benchmark 推荐数量参数不生效，让评测脚本按请求生成 top-k 结果 |