Field noteOct 22, 20261 min read1 source

Claude Mods: Flipping the Harness Inside Out (11 minute read)

In one minute

Latent Space describes Claude Mods as Claude Code plugins written in TypeScript or JavaScript.

What changed: Frames Claude Mods as an “Attention Interface” shifting the agent harness toward human-facing interaction, and argues the harness could become a platform for agent interfaces.

According to the article, a mod can respond to prompts, tool calls, and interface-render events; observe them; change their rendering; or handle an event instead of the default behavior.

The article says Claude Code ships six built-in mods, including agents-md, diff, plugin-authoring, sec-default, telemetry, and you-should-know. The last runs a side agent that surfaces information the main agent may have missed and is off by default. In the author's example, it flagged pre-existing uncommitted work across roughly 45 files.

The author also describes a custom Next Steps Supervisor mod that runs after a turn editing files, running commands, or making at least eight tool calls, then presents suggested next actions as clickable buttons. A separate open-source plugin, ModSmith, is presented as containing built-in mods and a template for composing new ones.

The source is an analysis and practitioner account. It does not independently establish the stability of the Mods API, the security properties of third-party plugins, broad availability, or performance and adoption data.

Evidence / sources

  1. 1latent.space/p/claude-mods
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicClaudeExplore the claude topic hub.
  2. 02topicClaude CodeExplore the claude-code topic hub.
  3. 03archiveField NotesOpen the latest editorial analysis.
  4. 04source ledgerSource LedgerInspect the public source evidence graph.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract