- Published on
Claude Code Becomes a Plugin Platform β How Mods Let Developers Redesign the Agent
Remember when smartphones became platforms, not just phones. Before the App Store, a phone did only what the manufacturer built in. After the App Store, anyone could redefine what a phone does. Something similar is happening to Claude Code.
What Mods Are
Claude Code 2.1.287, released October 1, 2026, introduced Mods β TypeScript functions that hook into Claude Code''s internal events. Developers can modify the agent''s behavior directly, at the machine level.
The scope of what''s possible:
- Prompt rewriting: Intercept and modify a prompt before it reaches the model
- Block or retry tool calls: Inspect a command before it runs and stop or retry it
- Automate permission decisions: Apply security policies without human prompts
- Redact secrets from tool output: Automatically mask sensitive values in logs or results
- Replace interface elements: Swap out default UI components with custom ones
Mods live inside plugins and install with a single /plugin install command.
Built-in Mods: Anthropic Eating Its Own Cooking
Anthropic converted several of its own built-in features into Mods at launch. The confirmed built-in set:
| Mod | Role |
|---|---|
/diff | Handles the diff panel UI |
sec-default | Applies default security policies |
telemetry | Manages usage data collection |
| Test framework | Plugin testing environment |
This matters because it means Anthropic itself manages Claude Code features through the same API it gives external developers. There''s no hidden internal layer β the playing field is level.
The Community: 1,423 Builds and Counting
The ClaudeMods.dev directory already lists 1,423 community builds. The most talked-about example is BlastRadius.
BlastRadius intercepts destructive shell commands β rm -rf, git reset --hard, force pushes β before they execute and shows a warning. It''s a safety net for the moments when an agent moves confidently in the wrong direction.
The key insight: anyone can build a BlastRadius for their specific risk profile and deploy it to their team. Company security policies become code. Every developer on the team runs the same rules automatically.
Security: What You Need to Know Before Installing
Mods launched with an explicit warning: Mods run unsandboxed, with the same machine privileges as Claude Code itself.
There is no technical barrier preventing a Mod from accessing the file system, making network requests, or reading secrets. Treat unknown plugins the way you would treat unknown npm packages β review the source code before installing.
The ClaudeMods.dev directory also carries sponsor content, so not every listed build is independently vetted.
Scenarios Mods Enable
Educational institutions: Deploy a Mod that prevents students'' Claude Code from modifying files outside the designated project directory. Accidental damage to lab environments becomes structurally impossible.
Enterprise dev teams: Auto-inject internal naming conventions into prompts, or flag code that imports deprecated libraries before the model suggests using them.
Individual developers: Install BlastRadius, then write a Mod that enforces linter runs before any commit. Your personal safety net, versioned and portable.
Sources: