Modes plugged into Claude Code: power in the kitchen, not the display
Plugin support is in public beta; the marketplace installs with one command and the community gathered 80-plus sub-agents. Not power, but maintenance six months out is the question.
Anthropic brought plugin support to Claude Code. The feature is in public beta. Plugins consist of slash commands, agents, MCP servers, and hooks. Teams pass their setups to each other this way. Plugins count as light packages. They switch on and off per job. Separately for debugging, testing, and deployment. Speed rises when idle ones close. Teams enforce code rules through plugins. They connect to internal systems over MCP. The framework ties packaging work to one shape. Modularity and collaboration stand together. So the power sits in the kitchen, not the display case. Setup trouble ends. Choosing trouble begins. The menu crowds. [1] [2]
My question is set. Who will maintain this six months from now. Installing a plugin takes a day. Keeping it alive takes every day. Every liked setup writes a maintenance debt. If nobody owns the debt, the order turns into a junkyard. Dotfiles existed before. Now marketplaces exist. The name changed, the trouble stayed. Ownerless settings rot. Rotten settings cause night outages. Every team learns this once through bitter experience. I hope someone keeps the ledger this time. Who keeps the ledger lives. Who skips it buries. Do not forget this.
Sharing is solved with marketplaces. Curated plugin collections sit in repositories. They install with a single command. A marketplace joins from GitHub or another source. Even the command name is known. Then you browse and install from the built-in menu. Community marketplaces are already emerging. Dan Avila published operations and docs plugins. Seth Hobson gathered more than 80 specialist sub-agents. Samples are ready too. Pull request reviews, security guidance, and SDK work. Setup is no longer the problem. Choosing is. The command is short. The menu is near. The choice is yours. [3] [4]
The numbers whet appetites too. Claude Code launched in May 2025. Its run-rate revenue passed 500 million dollars. It grew tenfold in three months. Enterprise developers embraced the product. The company overall reached a 5 billion dollar pace. A 13 billion dollar Series F stands behind it. The growth is real. But growth and maintenance are not the same thing. Where money arrives, order must arrive too. Otherwise every team builds its own marketplace. Then nobody understands anyone else's. Growth grows the team. Maintenance keeps it alive. They are separate jobs. [5] [6]
A single agent has two ceilings. Context windows are finite. Even millions of tokens are not free. One session cannot wear many hats. Architect, test author, and documenter do not fit at once. Sub-agents give three wins. Context isolation arrives. Each one sees only what it needs. Parallelism arrives. Independent jobs run together. Wall-clock time shrinks by one over N. Role specialization arrives. Each sub-agent gets its own skill, model, and hook. Model choice is written too. The frequently born run on Haiku. The main coordinator runs on Sonnet. Deep one-shot jobs run on Opus, then stop. Stopping is design too. [6] [7]
Let me write the counter-view too. Excess modularity breeds trouble of its own. A separate plugin per job means a separate version per plugin. Which of eighty sub-agents is current. Which is safe. Who audited it. As the marketplace grows, so grows the junk. Sometimes a single-file setup is more honest. Fewer parts mean fewer failures. Modularity is not free. The maintenance calendar pays its price. A team with no such calendar should not enter the marketplace. Ledger first, plugin later. Do not reverse this.
Let me end with an open question. Which of these marketplaces lives six months from now. Which turns into an abandoned depot. My watched measures are set. Not plugin counts, but update dates. Not star counts, but review records. I ask my team and myself the same question. Does the plugin you built live without you. If the answer is no, do not build it. If yes, write it, explain it, share it. What lasts is not the plugin but the maintenance habit. Write this on the wall. Read it every Monday. Report when the answer changes.
Source passages
- Claude Code Adds Plugin Support to Boost Developer Customisation ↗
Anthropic has rolled out plugin support in Claude Code, allowing developers to customise their coding environments through installable collections of slash commands, agents, MCP servers, and hooks. The feature, now in public beta, aims to simplify how developers extend Claude Code’s capabilities and share configurations across teams. Plugins act as lightweight packages that let users add or remove functions without affecting overall complexity. They can be enabled for
- Claude Code Adds Plugin Support to Boost Developer Customisation ↗
specific workflows, like debugging, testing, or deployment and turned off when not needed to optimise performance. According to Anthropic, the new framework standardises how developers package and share Claude Code customisations, combining modularity with ease of collaboration. The company says plugins can help teams enforce coding standards, provide custom automation tools, or connect internal systems through the Model Context Protocol (MCP). Framework authors and
- Claude Code Adds Plugin Support to Boost Developer Customisation ↗
technical leads can also bundle multiple configurations for specialised tasks. To help developers share their setups, Anthropic has introduced plugin marketplaces with repositories where curated collections of plugins can be hosted and installed using a single command. Users can add marketplaces from GitHub or other sources by running the command “/plugin marketplace add”, and then browse and install plugins through Claude Code’s built-in menu. Community-led marketplaces
- Claude Code Adds Plugin Support to Boost Developer Customisation ↗
are already emerging. The company highlights contributions by community engineers like Dan Ávila, who released plugins for DevOps automation and documentation. At the same time, Seth Hobson has curated more than 80 specialised sub-agents accessible directly via Claude Code. Anthropic has also released sample plugins for pull request reviews, security guidance, and SDK development. With plugin support now live, Claude Code is evolving into a modular AI development
- Claude Code Adds Plugin Support to Boost Developer Customisation ↗
platform, giving engineers greater control over how they integrate AI assistance into their coding workflows. Claude Code, launched in May 2025, has already surpassed $500 million in run-rate revenue, growing tenfold in three months. The product’s rapid adoption highlights its success among enterprise developers, contributing significantly to Anthropic’s overall $5 billion revenue run-rate following its $13 billion Series F funding round. ALSO READ: Databricks Launches
- Advanced Claude Code: Sub-Agents, MCP, Cost | CodeGateway ↗
engineering patterns FAQ Related reading Multi-agent orchestration: from one agent to a team topology Why spawn sub-agents at all? A single agent on a long task hits two intrinsic ceilings: context windows are finite (even at a million tokens, they aren't free), and one session struggles to wear multiple hats — architect, test author, doc writer — at once. Sub-agents (/agents) bring three structural wins: Context isolation. Each sub-agent only sees what it needs.
- Advanced Claude Code: Sub-Agents, MCP, Cost | CodeGateway ↗
Parallelism. Independent work runs concurrently, cutting wall time by 1/N. Role specialization. Each sub-agent can have its own Skills, model, and Hooks. Four topologies that actually show up Four ways to invoke Choosing models for sub-agents Sub-agents get spawned often. Default them to Haiku. Keep Sonnet on the main coordinator. Reserve Opus for one-shot deep reasoning (architectural reviews, tricky migration plans), then stop. MCP server integration: extending tools
