Skip to content
10/6 · Tue

Latest curated items

Today10/6Tue
  1. Hacker News · MCP78

    Flash-Agents: an MCP plugin that hands Claude Code's coding tasks to a DeepSeek Flash worker

    Flash-Agents is a Claude Code plugin that delegates bounded coding work—implementing slices, porting tests, reviewing diffs, mapping out a codebase—to a DeepSeek V4.1 Flash worker, while Claude keeps architecture, acceptance criteria, and final review.

    Why it matters: The author outsources Claude Code's coding tasks to a DeepSeek Flash worker and shares the sandbox, patches, and measured data, so you can judge the cost and safety boundaries for yourself.

  2. Hacker News · MCP78

    Spill：把超大的 MCP 返回结果移出上下文，存入本地 DuckDB

    Spill 是一个 Apache-2.0 开源工具，通过 Hook 拦截超过 32 KiB 的 MCP 工具返回，将其存为本地 DuckDB 表（~/.spill/spill.duckdb），智能体只拿到一个紧凑描述符，再用 SQL 查询而不是读入 5 万 token 的原始 JSON。

    Awaiting translation

    Why it matters: Spill 把超大 MCP 返回落到本地 DuckDB，让智能体用 SQL 取数，为上下文窗口紧张提供了一种可复用的思路。

10/5Mon
  1. Hacker News · MCP77

    我连续 37 天夜间测量 MCP 注册表：19.2% 的服务器 24 小时内改动了工具面

    作者搭建 mcp-transparency-log，按计划爬取官方 MCP 注册表中所有公开可达的服务器，记录其工具名、描述、JSON schema 和四个 annotation 提示，写入带签名树头的 append-only 日志。

    Awaiting translation

    Why it matters: 作者连续 37 天夜间爬取 MCP 官方注册表，用可复现的日志量化工具面变化，并公开了两次错误修正过程。

10/4Sun
  1. Hacker News · MCP76

    RugSnare: hash-pinning MCP tool descriptions to catch silent changes after approval

    RugSnare is a runtime integrity tool for MCP tool descriptions. It computes a normalized hash pin over each approved tool's { name, description, inputSchema }, and any silent change afterward triggers an alert and fails CI (exit 1).

    Why it matters: RugSnare hash-pins MCP tool descriptions, keeps watching for silent changes after approval, and shares measured data from 66 official server versions.

9/28Mon
  1. GitHubDaily78

    Paseo: one interface to manage Claude Code, Codex, and other AI agents

    The open-source project Paseo pulls command-line agents like Claude Code, Codex, OpenCode, and Pi into a single management interface. All agents still run locally on your own machine, and the project has already reached 18700+ stars.

    Why it matters: The original article shows how to bring multiple command-line agents into one interface and hand off tasks across agents, making it a useful reference for developers running several agents at once.

6/20Sat
  1. Kondasamy Jayaraman · Engineering Blog83

    Headroom：在智能体与账单之间做上下文压缩的本地代理

    Headroom 是一个开源、本地优先的代理，拦截 LLM API 调用并在请求到达服务商前重写 messages 数组，应用代码和模型都不用改。

    Awaiting translation

    Why it matters: Headroom 把上下文压缩放到代理层，读者可据此判断长会话智能体的 token 成本能否降下来。

4/24Fri
  1. Jesse Vincent66

    Prime Radiant 发布 Greenfield 与 Iterative Development 研究预览

    Prime Radiant 发布两款新技术的研究预览：Greenfield 把现有软件（代码库、文档、API 客户端等）转成行为规格语料，Iterative Development 则是一套基于 Superpowers 的智能体方法论，把大规格拆成需求、打包成开发 epic 后交给编码智能体实现。

    Awaiting translation

    Why it matters: Prime Radiant 公开两套工具的研究预览，读者可了解从旧代码库提取行为规格再驱动智能体重建产品的思路。

12/10Wed
  1. Jesse Vincent68

    packnplay: run a coding agent in a container with one command

    Jesse Vincent has released an open-source tool called packnplay. With a single command — `packnplay run claude --dangerously-skip-permissions` — it spins up a pre-configured throwaway container to run a coding agent.

    Why it matters: The author wraps all the tedious setup for running a coding agent in a container into one command, and also shares exactly how he handles credential conflicts with Claude Code.

You've reached the end