Tag
A user discovered that the DeepSeek Harness code integrates OAR (Open Agent Runtime) for implementing subagents. OAR is a vendor-neutral, unified TS API that can drive multiple coding agent runtimes such as Claude Code, Codex, Grok, and Kimi.
This article shares how to significantly improve quota efficiency and conserve your subscription allowance by configuring Codex’s subagent models (using GPT-6.1 Sol for the main work and delegating miscellaneous tasks to GPT-6 Luna), dividing responsibilities between scout/fixer subagents, and using AGENTS.md rules and prompt templates.
该文以面试题形式探讨了子 Agent 的生命周期管理、异步任务归属、异常恢复以及长时间 Shell 执行的阻塞控制等核心设计权衡。
The author dissects Claude Code's source code to explain how its multi-agent system layers the Agent tool, a task registry with a command queue, disk-based mailbox teammates, and a deterministic JS workflow engine — all funneled through a shared query executor with isolated contexts and message-only communication.
The content highlights the practical use of Qwen 3.8 27B as a subagent in AI orchestration setups, emphasizing its effectiveness alongside other models and tools.
A user shares a method to maximize Opus 5.5 AI model usage limits by using JEV to assign worker profiles and effort levels for task management.
The author experimented by having DeepSeek and Astra AI models work as subagents to build Mario Kart games, comparing the outputs and finding DeepSeek's version more soulful.
Lightning Orchestrator is an agent skill that uses fast subagents (SWE-1.7 Lightning) for implementation, achieving ~5x faster execution while keeping a frontier model as planner and reviewer. It works with Devin and supports parallel execution for larger tasks.
A practical tip from Simon Willison: allow Claude Code's Fable model to use its own judgment for testing and delegate coding tasks to subagents running cheaper models, reducing token usage while maintaining quality.
This is the sixth article in the series, explaining in detail the concept of subagent, its working principles, and its role in coding agents, including tool call and runtime mechanisms, as well as the applicable scenarios of different subagent types (fresh child, forked child, partial fork).
Microsoft released FastContext-1.0, a lightweight repository-exploration subagent for LLM coding agents that reduces main-agent token consumption by up to 60% while improving resolution rates by up to 5.5%.
A developer shares their workflow using Cursor's subagent harness with Opus 4.8 Max Thinking for long context understanding and implementing large features in Swift, emphasizing hands-on planning and phased acceptance testing.
Released physics-intern, a simple harness that significantly boosts the performance of reasoning models like Gemini 3.1 Pro on science problems, from 17.7 to 31.4, outperforming GPT 5.5 Pro.