AGENTIC COMMONSAI industry briefings

繁中EN

TopicAI Coding & Developer ToolsPublished

All English articles

Claude Code Mods Run Your Code Inside the Agent

Summary: On November 13, 2026, the ClaudeDevs account on X announced that you can build custom themes for Claude Code on desktop with mods — change the colors, add an animated background, or start from the theme-examples plugin. Per the mods overview docs, a mod is a plugin of JavaScript or TypeScript event handlers that runs inside Claude Code rather than beside it. The theme demo is the marketing; the underlying capability is broader, and so is the trust decision it asks you to make.

What mods can do that hooks can’t

Settings hooks, skills, and MCP servers run outside Claude Code. A mod runs inside it, so it can draw interactive panes with tabs, buttons, and text fields; restyle parts Claude Code draws itself, such as a tool call’s row or the spinner; hold or rewrite a tool call; run a /command with no Claude turn; and share state between handlers. Mods work in the CLI, the desktop app’s Code tab, and the VS Code extension. They install as plugins from a marketplace, and Anthropic ships samples in the claude-code-playground repo: token-weather (a context-window forecast above the prompt), blast-radius (holds a risky command like rm -rf or a force push and shows what it would change), and replay-theater. You can also just describe a mod to Claude and have it write one.

The part to read before you install

A mod runs with your permissions: it can read and write your files, start processes, make network requests, read environment variables and API keys, see and rewrite your session, spend your usage, and approve tool calls before you’re asked. The docs are explicit that mods aren’t sandboxed. Two limits help: with the cc-plugin-sec-default guard loaded, a user’s mod by default can’t approve tool calls your deny rules would stop, and a mod can’t restyle the permission prompt. Before loading anything, claude plugin validate lists a mod’s hooks and calls without running it. Mods need Claude Code v2.1.287 in the terminal or v2.1.286 in the desktop app.

Our take: Act now, but only on mods you can read. Cloning a sample and loading it with --plugin-dir for one session is a cheap, reversible afternoon, and blast-radius is a real guardrail, not a toy. Treat a third-party install as running a stranger’s script as your user. Read-only users in throwaway sandboxes can skip this. If Anthropic closes the guard’s exceptions and mods stay unable to touch the permission prompt, we would revisit third-party installs — and remember the runtime changes under you (see our Claude Code quality postmortem).

Sources

Original post
AGENTIC COMMONSOperated by PHLEGON LABS
ShareXEmail
Support us

Related reading

  1. Bun 1.4 Is Rust Now: 64 Claude Agents Ported It in 11 Days

    Bun v1.4.0 is the first Rust-written release of the JS runtime: Claude Code workflows plus Claude Fable 5 ported 535k lines of Zig in 11 days with 64 agents, for about $165k.

    Open Source

  2. Anthropic Postmortem: Why Claude Code Got Worse for a Month

    Anthropic's April 23 postmortem ties a month of Claude Code complaints to three changes: lower reasoning effort, a cache bug wiping thinking history, and a verbosity prompt rule.

    Anthropic

  3. 16 Claude Agents Built a C Compiler: 100K Lines, $20K

    Anthropic researcher Nicholas Carlini ran 16 parallel Claude Opus 4.6 agents over ~2,000 Claude Code sessions to write a 100,000-line Rust C compiler that boots Linux 6.9 — for under $20,000.

    Anthropic