Seven Claude Code Tools Shipped This Week
What developers shipped this week.
Seven Claude Code Tools Shipped This Week
The Shiplog
cc-switch (GitHub) - A cross-platform desktop assistant written in Rust that aggregates Claude Code, Codex, OpenCode, OpenClaw and Gemini CLI into a single interface, with 98,108 stars as reported by the source. The appeal is obvious: developers who move between AI coding tools spend real time context-switching between interfaces, and this removes that friction. The star count is unusually high for a tool this specific, which warrants some scepticism about organic growth, but the underlying problem it solves is genuine. Worth evaluating if you operate across multiple AI coding environments.
claude-mem (GitHub) - A TypeScript project that provides persistent context across Claude agent sessions, capturing what an agent does during a session so that context survives between runs. This addresses one of the most practical limitations of working with Claude agents: the fact that each session starts cold. Memory and continuity are not solved problems in agentic AI, and this is a community attempt to patch that gap. Developers building multi-step or long-running agents with Claude should look at this before rolling their own solution.
caveman (GitHub) - A Claude Code skill written in JavaScript that claims to cut token usage by 65% by compressing how the model communicates during a session, with 71,128 stars as reported by the source. Token costs are a real operational concern for teams running Claude Code at any volume, and a 65% reduction would be material. The name is deliberately absurd but the premise is sound: verbose internal reasoning costs money, and constraining it does not necessarily reduce output quality. Test it on a real project before committing.
planning-with-files (GitHub) - A Python project with 23,040 stars that provides persistent, file-based planning for long-running agentic tasks, using crash-proof markdown plans to maintain state across agent runs. State management is the unglamorous infrastructure problem that breaks most agentic systems in production, and markdown files are a deliberately simple, inspectable solution. The deterministic approach means a human can read and edit the plan at any point, which matters when something goes wrong. Developers building autonomous agents who have been burned by opaque state failures will find this approach immediately familiar.
awesome-claude-code-subagents (GitHub) - A curated collection of over 100 specialised Claude Code subagents covering a wide range of development use cases, written in Shell and maintained by VoltAgent, with 21,326 stars. Subagents allow Claude Code to delegate specific tasks to purpose-built agents rather than handling everything in a single context, which improves both reliability and cost. This is essentially a pattern library: not a framework to install but a reference for how others have structured real agent specialisation. If you are designing a multi-agent Claude Code system, start here before writing anything from scratch.
The Take
There is a pattern across this week's Claude Code releases that is worth naming directly. The tools that are attracting the most attention are not new capabilities. They are infrastructure for the capabilities that already exist.
Persistent memory, token compression, crash-proof state management, subagent pattern libraries: none of these extend what Claude can do. They make what Claude already does survivable in production. That is a meaningful distinction.
Most AI tooling coverage focuses on what models can now do that they could not do before. But the actual bottleneck for teams shipping production software with Claude Code is not capability. It is reliability, cost and continuity across sessions. The community appears to have worked this out ahead of the tooling vendors. The repositories shipping this week are filling gaps that Anthropic has not yet addressed in the core product.
That is either a sign that the ecosystem is healthy and self-correcting, or a sign that the foundation still has structural gaps that should have been closed before developers were encouraged to build on it. Probably both.
Is the Claude Code ecosystem building genuine infrastructure, or papering over gaps that Anthropic should have shipped by now? Tell me in the comments.
Forward this to one person who should be using AI better than they are. Reply with what you built, tried or broke this week. I read every one.
Gareth, founder of The Anthropic Stack (theanthropicstack.com)