On this page
Claude Code’s ecosystem has grown fast over the past few months, evolving from a simple CLI tool into a complete platform with Skills, Hooks, Agents, and Subagents. Here’s the current landscape.
Want AI working in your own job, not just in demos? See the 3-month 1-on-1 program →
Skills — Now Supported Everywhere
Skills are no longer a Claude Code exclusive. Codex supports them, and even Gemini does now. But the ecosystem maturity gap is still wide — Claude Code’s Skill ecosystem is the most active by far, with community contributions far outpacing other platforms.
The community’s creativity is entertaining too. Someone built a Skill called moyu that converts your own code into spaghetti before you resign — “gone but never forgotten.” Another person dreamed up using /schedule to auto-generate fake weekend stories every Monday morning and fake achievements every day before work, sent straight to your inbox for a one-minute commute read before you bluff your coworkers.
Jokes aside, these cases reflect Skills’ core value: encapsulating repetitive workflows into a single command, whether the content is serious or absurd.
Hooks — The Gatekeeper That Intercepts Commands
Hooks are more enforceable than CLAUDE.md. CLAUDE.md is advisory — the model can choose to ignore it. Hooks operate at the system level and can’t be bypassed.
Someone in the community built a vague-prompt-detection hook that intercepts every user command and evaluates whether it’s too vague. Clear prompts pass through; vague ones trigger a Skill that asks 1-6 clarifying questions. Useful for anyone who tends to be hand-wavy with instructions.
Agent View — Multi-Session Management
Claude Code’s Agent View feature lets you manage multiple sessions collaborating on a single task. Upgrade to the latest CLI and run claude agents to start.
This solves a longstanding pain point: you no longer need separate terminal windows for different agents. There’s now a unified interface for coordination.
Fork Subagent — Inherit or Isolate
The experimental fork subagent feature lets child agents inherit the parent agent’s context. A related experiment, Agent Team, has the main agent act as a manager dispatching tasks to team members.
One gotcha after enabling fork: you need to explicitly tell the main agent whether to inherit context or not. Otherwise it defaults to forked agents with full context inheritance — but sometimes for auditing you specifically don’t want the same brain grading its own work. An isolated, context-free review is often more valuable.
What the Ecosystem Means
Each of these features is a small tool on its own, but combined they form a complete workflow platform. Skills encapsulate processes, Hooks enforce rules, Agents divide labor, and Subagents decide between isolation and inheritance.
For heavy daily Claude Code users, these aren’t nice-to-haves. They’re infrastructure.
Postscript (2026-10-20): Fork subagents aren’t experimental anymore. Since v2.1.232 (August 2026) forking is on by default, so the “say explicitly whether to inherit context” gotcha now applies to everyone. To turn forking off, set
CLAUDE_CODE_FORK_SUBAGENT=0.