Claude Code 2.1.300: what changes for mods?

virtualText, prompt.change, and prompt.submit do different jobs

The official 2.1.300 release notes introduce virtualText in prompt decorations. It draws text in the input area without adding it to the draft or submitted prompt. An editorial use case is a hint you want the person to see without making it part of their request. This feature alone does not establish token savings for a whole mod.

The new prompt.change hook covers input-box changes such as history recall, clearing, undo, and a plugin filling the box. The release also fixes prompt.edit missing a newline entered through the newline keybinding. If your feature tracks draft state, our recommendation is to include those paths in its regression checks.

Need Claude to read extra instructions?

Use the separately documented prompt.submit mechanism when you intend to change the request. The events guide describes rewriting e.text, appending model-visible context, or returning a drop reason before forwarding the prompt. Added context can leave the visible message unchanged while giving Claude additional text.

Choose explicitly: a reminder for the person, a reaction to draft changes, or an instruction for the model. A visible hint should not be mistaken for an instruction Claude received. Conversely, an unchanged visible message does not establish that no extra context was attached.

Agent transcript navigation and workflow-agent rewriting

The release adds $.ui.view({ agentId }) to show an agent’s transcript and $.ui.view({}) to return. It also allows agent.spawn to rewrite a workflow agent’s prompt and model; previously that workflow-agent path could only refuse it.

Our suggested distinction for dashboard authors: opening a transcript is a navigation feature; changing a spawned agent’s instructions changes its behavior. Explain those controls separately in your README so a reader can tell what a button will do before using it. Neither addition proves that an existing community dashboard already supports the new behavior.

Review a mod before relying on the new APIs

1. Check the session’s version with /status. General Mods support and the specific additions in 2.1.300 have different version requirements. Use the installation and display checklist if a mod loads without visible UI.
2. Read the author’s requirements and release notes for the revision you plan to use. Do not assume every mod needs an update merely because Claude Code has a new release.
3. For draft-aware features, review typing, newline, history recall, clear, undo, plugin fill, and submission as separate cases. These are proposed checks, not results we measured.
4. Review what the mod can access using the pre-install capability checklist. Display-only text describes that text’s behavior, not the capabilities of the entire plugin.

Why no copy-and-paste implementation here?

At our October 15 review, the official event reference still listed prompt.edit but did not document prompt.change or the new decoration’s full shape. The release notes confirm the additions, but do not supply a complete implementation contract. Check documentation and the types for your installed build before writing a handler; we do not infer a signature from a feature name.

The interface guide remains the starting point for panes, bands, controls, and rendering. A prompt-box decoration is one design choice among those surfaces. Pick the least intrusive surface that makes your information useful.

Original sources and related guides

Anthropic’s v2.1.300 release · React to events · Prompt event reference · Draw in the interface

New to customization? Compare mods, plugins, skills, hooks, and MCP, or browse workflow mods. Source links and editorial review do not mean execution testing or a security audit.