ClawTab vs SkillClaw

ClawTab is a macOS desktop app that turns managing multiple AI coding agents from a terminal circus into an organized workflow. Built by indie developer Tõnis Tiganik, it provides a proper GUI for running Claude Code, Codex CLI, and OpenCode in parallel — with a sidebar showing per-agent status, pane splitting, auto-yes passthrough, and the ability to trigger agent restarts from your phone. The core problem it solves: once you start running more than 3-4 coding agents simultaneously, tmux panes become unreadable and you start losing context on which agent is doing what. ClawTab gives each agent a labeled tab with status indicators, scrollable history, and the ability to quickly switch contexts without losing your place. It's the kind of tool that only makes sense in a world where shipping a feature means spinning up 10 agents on 10 tasks at once — and that world is arriving fast. Version 1.0 launched on Product Hunt today and is already getting traction from the vibe-coding crowd.

SkillClaw is a research framework from Alibaba's AMAP-ML team that enables collective skill evolution for LLM agent systems deployed at scale. The core idea: instead of each user's agent interactions existing in isolation, SkillClaw aggregates anonymized skill-improvement signals across all users to continuously refine a shared library of reusable agent skills — without requiring centralized fine-tuning. The framework introduces a three-component architecture: a Skill Extractor that identifies and catalogs atomic capabilities from interactions, a Skill Evolver that proposes improvements based on aggregate feedback, and a Skill Selector that routes tasks to the best-available skill version per user context. Published on April 9 and hitting #1 on Hugging Face trending papers this week with 277 upvotes, the paper reports significant improvements over per-user baselines on complex multi-step agentic tasks. This matters especially for production agent deployments where cold-start problems are severe — a new user's agent immediately benefits from millions of prior interactions. It's a fundamentally different model of agent improvement than either fine-tuning (expensive, periodic) or RAG (retrieval-only, no learning).