Tag:architecture

All the articles with the tag "architecture".

• Claude Code doesn't trust Claude with permissions

  How Claude Code enforces permissions with deterministic code, not model inference. The one LLM-based path is internal-only, fail-closed, and stripped from the public package.

• Claude Code is a vibe-coded mess. Some of it is actually good.

  Some ideas worth stealing from the leaked Claude Code source: deferred tool loading, diminishing returns detection, cache-aware context compaction, and coalesced background extraction.

• LangGraph's HITL Has a Double Execution Problem

  LangGraph's interrupt/resume primitives are genuinely clean. But the checkpoint model has a side effect that silently double-executes tools, and approval binding is positional rather than explicit. Here's what that means in practice.

• Stop Using Chat History As Your Agent's State Store

  The chat interface is fine for users. The flat message model behind it erases the semantics that agentic systems actually need to operate correctly.