Claude Code Mods Turn Guardrails Into Installable Code

Anthropic added mods to Claude Code on October 1, 2026: small TypeScript functions that change how the tool behaves. A mod can rewrite prompts, add interface elements, replace built-in features or add new ones. They hook into events across a session, including tool calls, permission requests and screen rendering, and they work in both the CLI and the desktop app.

The detail that matters is that Anthropic’s own features run on the same mechanism. The /diff command now runs as a mod, so it can be disabled or replaced. That makes the extension point real rather than decorative. Named uses in the post include rewriting prompts, blocking tool calls, managing permissions and redacting secrets.

Mods ship inside plugins, and Anthropic has just opened the door for outside builders. A submission portal now lets developers on paid plans publish plugins that bundle MCP connectors, Agent Skills, or both, with automatic validation and safety scanning on submission. Put the two together and a consulting team can package a per-engagement set of guardrails as something installable rather than a document nobody reads.

Team and Enterprise plans get a built-in security mod, sec-default, that loads first to prevent risky modifications. Organizations can write their own mods for CI/CD integration, production safeguards and audit logging.

For anyone running Claude Code on client work, the practical move is small. Start with the guardrails you already enforce by hand: which commands need approval, which files never leave the machine, which secrets get redacted. Turn each into a mod, test it on one engagement, and keep the ones that catch something. Governance that ships as code gets used. Governance that ships as a policy PDF gets skimmed once.