Learn from the past to understand the present
This article maps cognitive psychology's semantic, episodic, and procedural memory to AI agent architecture, then details Deep Agents' MemoryMiddleware implementation that loads AGENTS.md files at initialization and injects their content into system prompts, functioning as procedural memory with semantic elements.
This article analyzes Deep Agents' Skills implementation, revealing a three-layer progressive disclosure loading pattern where metadata loads at startup, full instructions load on activation, and resources load on demand. It explains how before_agent and wrap_model_call hooks inject skill metadata into system prompts, and why SkillsMiddleware positioning differs between main agents (early for coordination) and sub-agents (late for separation of concerns).
This article details Deep Agents' context management implementation using SummarizationMiddleware, which automatically compresses conversation history during model calls by offloading old messages to backend storage and replacing them with generated summaries, including threshold configuration, cutoff determination, and the three message views that enable traceable, iterative summarization.
This article dissects Deep Agents' SubAgentMiddleware implementation, revealing how the task tool dynamically delegates complex work to isolated sub-agents with separate context windows, detailing dynamic tool construction, state isolation via excluded keys, system prompt injection for delegation guidance, and a Deep Research case study showing research-agent delegation in action.
This article explains the SO_LINGER socket option's three closure modes, demonstrates why SO_LINGER=0 sends RST instead of FIN, and outlines Linux's preferred solutions for TIME_WAIT accumulation—connection reuse, tcp_tw_reuse, tcp_fin_timeout, and port range expansion—over abrupt connection termination.
This article dissects Deep Agents' shell command execution via the FilesystemMiddleware's execute tool, detailing its parameter validation, backend resolution through SandboxBackendProtocol, LocalShellBackend's subprocess-based implementation, BaseSandbox's abstract interface for isolated sandboxes, and a practical example using a local workspace.
This article analyzes Google's BBR congestion control algorithm, comparing its model-driven approach with CUBIC's loss-based design, explains why BBR hasn't become Linux's default despite advantages in high-BDP and wireless scenarios, details BBRv2/v3 improvements addressing fairness and robustness, and examines the Linux kernel's congestion control framework implementation.
This article dissects Deep Agents' FilesystemMiddleware implementation, covering six filesystem tools (ls, read_file, write_file, edit_file, glob, grep), two core middleware hooks for system prompt injection and large-result eviction, and the BackendProtocol layer with four backend implementations including CompositeBackend's longest-prefix routing for complex scenarios.
This article explores why Linux separates socket listening into bind and listen system calls, analyzing the design rationale, kernel internals, port allocation strategies, and connection queue behaviors through experiments and Linux 5.15 source code analysis.
LangChain's Deep Agents framework provides a batteries-included architecture for building complex LLM agents with built-in planning, filesystem access, shell execution, sub-agent delegation, smart default prompts, and context management, enabling long-running, multi-step tasks like coding and research.
