Claude Mods: Flipping the Harness Inside Out

What it claims

Claude Mods are Claude Code plugins made of TypeScript or JavaScript functions that act as runtime middleware for the harness. The author argues they make the human-facing layer, which the author calls the "Attention Interface," programmable by the user. Anthropic already builds parts of Claude Code as mods. Six ship built in. The author's own Next Steps Supervisor, in the open-source ModSmith plugin, ran on 99 of 111 turns, and its README estimates about 5-6k input-equivalent tokens per check.

How it works

A mod subscribes to events such as a submitted prompt, a tool call, an interface render or the end of a turn. For each event it can observe it, change how it renders, or handle it in place of the default behavior. Claude Code has a built-in skill, plugin-authoring, that lets Claude write a mod from a plain-language request.

The six built-in mods are listed under /plugin, then Installed, then Built-in:

• agents-md: loads AGENTS.md as project instructions.
• diff: takes over /diff and draws its pane.
• plugin-authoring: the mod-writing skill.
• sec-default: protects organization-managed settings from user-installed mods.
• telemetry: sends analytics records logged by Claude Code and built-in mods.
• you-should-know: a side agent that posts a note above the prompt about things the user may have missed. It is the only one off by default.

In the author's example, You Should Know flagged that the file being edited, along with about 45 others, held uncommitted earlier work. It suggested a commit or stash before discarding the demo edits.

Next Steps Supervisor triggers after any turn that edits files, runs commands or makes 8 or more tool calls. It forks the conversation, re-reads it, and lists gaps and shortcuts. It then offers a suggested next step as a clickable button that sends the instruction to Claude Code. ModSmith also includes a template for Claude to build new mods that work together.

Evidence

The evidence is one user's experience, not a benchmark. From October 15, 99 of the author's 111 Claude Code turns met the supervisor's trigger. In one session the supervisor marked a task "Solved" while also listing two gaps and three shortcuts.

How to use it

Enable You Should Know from the built-in list in /plugin. To turn off the supervisor for a session:

/supervisor off

The post gives no install command for ModSmith.

Limits

• Noise. The supervisor's trigger fires on almost every turn, so the author began skimming its output.
• Cost. Each check costs about 5-6k input-equivalent tokens on a 50k-token session when served from the prompt cache. It costs more once the cache expires.
• Trust. Anthropic's docs state that "mods aren't sandboxed." A loaded mod can read and write files, see every prompt and spend usage. Built-in mods keep running under --safe-mode. The author reads third-party mod code before installing it.

The comparison with Cursor's PR Routing & Approval notes one design detail. If a PR edits an approval policy, Cursor reads the policy from the base branch, so the PR cannot relax its own review.