The Good Parts of Claude Code

The Case for Feature Parity

While the creator expresses significant frustration with Anthropic's restrictive subscription models and desktop application, this analysis focuses on specific architectural and UX patterns in Claude Code that outperform competitors. The goal is not to promote Claude Code exclusively, but to highlight features that should be adopted as industry standards for agentic coding harnesses.

Scriptable Skills and Execution

Claude Code’s approach to "skills" distinguishes itself by allowing the agent to execute scripts directly within the skill definition. Unlike static markdown-based skills, this allows the model to perform pre-execution tasks—such as checking a local cache directory or listing repository contents—before the primary agent loop begins. This reduces context window noise and improves reliability by offloading logic to the file system rather than forcing the LLM to "reason" through basic file discovery.

Context Management via Imports

Anthropic’s implementation of claude.md allows for recursive file imports, enabling users to modularize their instructions. By using @path/import syntax, developers can pull in existing documentation (like README.md or package.json) or even link their existing agents.md files without resorting to symlinks. The addition of claude.local.md provides a clean override pattern, allowing individual developers to maintain personal preferences without polluting shared repository configurations.

Code-Driven Workflows

Workflows represent the most advanced feature in Claude Code. Rather than relying on rigid, pre-defined tool calls, the agent writes actual JavaScript code to orchestrate sub-agents. This allows for dynamic, multi-phase execution (e.g., audit, rule, and verify phases) that can be tailored to the specific task. By writing code to filter data programmatically before it enters the context window, the agent avoids the common pitfall of flooding the LLM with excessive data from MCP servers. However, this power comes with a significant cost, as parallelized sub-agent workflows can consume tokens rapidly.

Terminal UX and Performance

For power users, the "full screen" (alt-screen) rendering mode provides a cleaner experience than standard terminal re-rendering, effectively isolating the agent's output from the user's scroll buffer. While this can complicate SSH/tmux workflows, it offers a more stable environment for long-running agentic tasks.