Tag
A maker shares early progress on a workspace where multiple coding agents (Claude Code, pi, opencode) collaborate in shared rooms, discussing design, debating, changing positions, and splitting work autonomously while awaiting human approval. They are looking for early feedback from people who run multiple agents.
Veynod is an open-source agent coding runtime where coding agents claim tasks from a shared board with leases, evidence, and review gates; a major refactor unifies the Runtime and Kernel into one system with a MIT-licensed bridge layer. The scheduler hot path is written in Rust via PyO3, with Postgres as the durability fence and a deterministic kernel reachable only through 12 typed capabilities.
A developer is building 'lineup', an open-source CLI that summarizes what coding agents changed and flags risky edits (deleted tests, config/env changes, auth code, new dependencies) before they become a PR. The post also asks the community what edits they'd most want flagged and how they currently review agent changes.
DeepSeek released dsh-superagent-prototype, an open-source proactive coding agent that coordinates multiple agents to handle GitHub tasks, reviews, and CI feedback in parallel with minimal human oversight.
A developer shares an OpenCode setup where GLM-5.3-Flash running on H200s handles most coding tasks while GPT-6.1 Sol acts as the orchestrator, highlighting a multi-model agentic workflow.
The open-source tool ai-design-check breaks down the "AI look" into 70+ countable check rules (spacing grid, type scale, gradient buttons, emoji icons, 'Welcome back' greetings, etc.). It's zero-dependency, requires no AI or API key, runs on React/Vue/Svelte/HTML/SwiftUI/Jetpack Compose, and can be hooked into Claude Code so the agent automatically fixes design issues itself.
An essay examining 'cognitive surrender' — the seductive but risky habit of uncritically accepting AI agent output — and arguing for a conscious balance of trust when working with coding agents to avoid technical debt and lost system familiarity.
CrossSWE introduces a multi-repository benchmark of 171 real-world instances showing that current LLM coding agents struggle with cross-repository consistency, with the strongest model (GPT-6.1 Sol) reaching only 58.48% Pass@1 and cross-repository inconsistencies accounting for 44.2% of failures.
SWE-Trail is a history-guided trajectory collector for coding agents that uses earlier rollouts to steer where collection effort is spent without exposing that history to the agent, avoiding train-test mismatch. Trained on its trajectories, a 124B Ling-3.0-flash-base student gains 2.33 and 1.60 percentage points in resolution on SWE-bench Multilingual and SWE-bench Pro over independent sampling.
DeepSeek's Harness team open-sourced dsh-superagent, a research prototype of the underlying architecture for their always-on autonomous Super-agent, packaged as a proactive coding agent for a single GitHub repository. The MIT-licensed release listens to GitHub and Feishu, handles work items in isolated agent sessions, and is designed to minimize human attention while maximizing delivered value.
Kanzo is an open-source local kanban board and MCP server that lets developers and coding agents like Claude Code or Codex share the same task board, tracking agent progress without copying tasks between tools.
GPT-5.6 Sol, running in Codex with low thinking, autonomously completed a peaceful-mode Factorio run in 13.5 in-game hours, launching a rocket via a Python game-manager library that mirrors human GUI constraints, marking a clear improvement over previous GPT-5.5 attempts.
AgentMash 是一款开源工具，为团队的编码智能体提供多人协作的 AI 体验，支持多个 coding agent 同时协同工作。
Backdoored builds of the Anthropic.claude-code extension briefly appeared in Cursor's marketplace, potentially reaching up to 5,000 users via auto-update, installing SSH backdoors and persistence mechanisms that survive simple updates or uninstallation.
This article introduces Obelisk, an open-source system that indexes historical sessions from coding agents like Claude Code and Codex into SQLite for direct retrieval by agents. Its design is inspired by the paper "Is Complex Agentic Memory Always Needed?", which found that a simple RAW-RAG approach (retaining raw conversation history and retrieving on demand) outperformed complex agentic memory schemes across multiple memory benchmarks.
Recommend Stanford’s free course CS146S, “The Modern Software Developer” (taught by Mihail Eric), and introduce the accompanying open-source project NamiAgent—a Chinese-language tutorial on building an AI agent from scratch, progressively implementing core capabilities such as tools, permissions, context, Skills, MCP, sub-agents, and memory, complete with web-based visual debugging.
论文提出 DecompGym 基准，包含来自九类问题族的 81 个实例，用于评估 12 个 GPT、Claude、Gemini 编码智能体设计与实现 Benders 分解和列生成求解器的能力，并比较了独立重生成与反思两种失败修复策略的效果。
The author explores what to do when AI coding agents produce code faster than a team can comprehend it, and how five layers of governance — compiler-enforced constraints, architectural skeletons, read-only contracts, refactoring loops, and disposable implementations — can counteract system entropy and human cognitive bankruptcy to preserve software maintainability.
The tweet shared Hugging Face’s tutorial “A curriculum for post-training agents,” which centers on coding agents and covers SFT, distillation, GRPO, OpenEnv multi-turn interactions, and training with a real harness, using experiments to clearly demonstrate the risks of reward hacking.
swyx 宣布 AI Engineer Code Summit 2026 将于 11 月 10-12 日在旧金山举办，聚焦这一年 agentic coding 的进展，汇集顶尖 AI 工程师、研究者与创始人，涵盖 coding agents、代码审查、评估与生产级基础设施等主题。AI Engineer Code Summit 2026 returns Nov 10-12 in SF, gathering top AI engineers to cover a year of agentic coding, coding agents, internal agents, evaluations, and production infrastructure.