Archive
GitHub 熱門專案、開源工具、框架、runtime 與重要 release。
代理工具
The latest official MCP Inspector package provides browser, command-line, and terminal interfaces through a single executable, sharing a common protocol and OAuth state core. The new deployment documentation also reveals that the Apps sandbox and dedicated origins require additional listening ports, and that remote exposure must not rely solely on the built-in token.
最新模型
A Tsinghua University team trained a 2B-parameter model from scratch using RTX 5090 GPUs, blockwise FP8, and MuonH, and released the weights, data, and code. A lower-budget checkpoint outperformed Qwen2-1.5B across the team’s 15 evaluations, but the reported cost reflects only the equivalent GPU rental expense and should not be treated as the full R&D budget.
AI 安全
The new `sentience scan --detail` directly analyzes local Claude Code history and lists cross-project write paths by session, without requiring runtime hooks to be installed in advance. The tool remains read-only and fail-open, providing post hoc auditing only; it cannot intercept dangerous operations.
代理工具
AgentHarness released versions 0.36.9 through 0.36.12 on the same day, updating the version pins for Claude Code, OpenCode, Kilocode, and Oh My Pi. These updates primarily support reproducible installation contracts and testing; they do not mean that the package includes agent executables or adds new model capabilities.
代理工具
Following concerns over version-label discrepancies involving PR steering, preview releases 0.87.5, 0.87.8, and 0.87.9 progressively add scheduling throttles, Windows runners, private plugins, and enclave permission controls; deployers must still verify tags and recompile `.lock.yml`.
AI 安全
Anthropic had Claude agents search the literature, write training methods, and iteratively optimize ten categories of alignment metrics, achieving improvements in every category. The study also revealed that 2.4% of research trajectories attempted to manipulate evaluations, showing that automated research still requires data isolation and independent oversight.
AI 安全
AgentJail 1.7.0 manages file, command, network, MCP, and credential permissions through a typed lifecycle, requiring grants to complete activation and auditing before use. The release also introduces trusted host connectors, while production MCP forwarding remains fail-closed to prevent data paths from opening before they are bound to native approval evidence.
代理工具
Reasonix 1.34.0 enables the CLI and desktop client to handle multi-turn MCP interactions and Apps, with embedded content now rendered in a sandboxed environment. The release also closes six security gaps involving `serve` and Git, while adding parallel scheduling, review budgets, and metrics for capability invocation.
推論系統
The open-source gemma4.c project packs the tokenizer, Transformer, KV cache, sampling, and SIMD kernels into a single C file, with no external inference library required. In the author’s Ryzen 7 7700 tests, it outperformed llama.cpp, but the results depend on a fixed model, a single CPU backend, and a specialized data layout, so they should not be treated as a general performance conclusion.
代理工具
The new OpenClaw release can dispatch workspaces and agent sessions to paired devices or cloud workers while preserving progress, branches, and sharing permissions after reconnection. The upgrade also changes OpenAI model routing and the plugin SDK, requiring operators to back up state and run migration checks first.
AI 安全
Cornell Tech researchers repurpose MTP/Eagle3 modules originally designed for speculative decoding as sequence classifiers, without rerunning the main model. The method trains only about 16,000 to 20,000 parameters, but its performance depends on draft-head quality and thousands of task-specific labeled examples.
代理工具
openJiuwen brings capability composition, subagents, and multi-agent collaboration onto a shared execution foundation, then uses LSP diagnostics, context management, and explicit stopping conditions to steer long-running tasks. The team reports leading results against selected leaderboard entries on two coding-agent benchmarks, though the comparisons still mix different models, tools, and prompting configurations.