· 8 min · News

Claude Code Features Roundup: TypeScript Mods, Draft Recovery and Plugin Hooks (October 2026)

Claude Code Features Roundup: TypeScript Mods, Draft Recovery and Plugin Hooks (October 2026)

The latest Claude Code features bring TypeScript mods, recovery for accidentally cleared prompts, and new plugin hooks for working with agents. This October 4, 2026 roundup covers releases 2.1.287 through 2.1.289, explains the commands and configuration changes developers can use, and highlights the permissions and compatibility details worth checking before adopting mods.

Claude Code features at a glance

The first three October releases combine a substantial extension feature with smaller improvements to everyday development. Claude Code 2.1.287 introduced mods; 2.1.288 added draft recovery and review controls; 2.1.289 expanded plugin-hook support and addressed mod loading and permission problems.

The table below gives you a starting point based on the workflow you want to improve.

Release Date Main additions and fixes Where to focus
2.1.287 October 1 TypeScript mods, “You should know,” session-name filtering, MCP URL prompts Plugin customization and MCP compatibility
2.1.288 October 2 Cleared-prompt recovery, review findings limits, mod text selection, session shortcuts Prompt editing, code review, and session navigation
2.1.289 October 3 Teammate spawn hooks, consistent agent IDs, agent states, mod and permission fixes Agent-aware plugins and managed-machine permissions

These releases address different parts of the same workflow. Mods let plugins change Claude Code’s behavior, while draft recovery makes prompt editing more forgiving. The subsequent hook and permission fixes matter when those extensions interact with agents, tools, and organization-managed settings.

TypeScript mods extend plugin behavior

Claude Code 2.1.287 introduced Claude Mods with the release-note description “plugins may now modify deeper behavior.” Mods are small TypeScript functions packaged inside plugins, and they work in both the CLI and desktop app.

The October 1 mods announcement describes several capabilities: rewriting prompts, adding UI, replacing built-in features, and blocking, rewriting, or retrying tool calls. That scope makes mods relevant when you need to change how Claude Code behaves during a task.

For developers familiar with hooks, the useful distinction is the scope explicitly documented for mods: they can alter prompts, UI, built-in behavior, and tool calls. The Claude Code hooks guide provides a reference point when evaluating which extension mechanism fits a project.

Creating and testing a mod

The versioned 2.1.287 mod README documents an exported register(on, options) function and handlers shaped ($, e, next). The plugin layout includes .claude-plugin/plugin.json, hooks/hooks.json, and TypeScript files under hooks/.

The documented development commands are:

/plugin-types
claude --plugin-dir mods/diff
claude plugin test mods/diff

/plugin-types writes declarations. The other two commands are the snapshot’s examples for running Claude Code with the example plugin directory and testing that plugin.

The snapshot also warns that the API may change between releases without notice. Treat the versioned documentation as part of your implementation reference, especially when upgrading a plugin that depends on mod APIs.

Access, trust, and the built-in mod

Mods run with Claude Code’s access to the machine and are not sandboxed. Anthropic’s announcement advises installing them only from trusted sources. Review the source and origin of a mod before enabling it, particularly when it can rewrite or retry tool calls.

Release 2.1.287 also added “You should know,” a built-in mod whose side agent flags things that the user or Claude might miss. Its enable command is:

/plugin enable cc-plugin-you-should-know@builtin

Availability is stated as first-party sessions with telemetry enabled. That eligibility detail matters: the announcement does not establish availability in every session configuration.

Draft recovery and code-review controls

Claude Code 2.1.288 adds two useful controls for interactive work: recovering a prompt cleared with Ctrl+C and choosing how many code-review findings to report.

Restore a prompt cleared with Ctrl+C

If Ctrl+C clears your prompt, press Up while the prompt is empty to restore the draft. Recovery includes pasted text and images.

The release describes a specific recovery action for a cleared prompt. It does not establish a general draft archive or recovery behavior for every way a session might end. The sequence to remember is simply: Ctrl+C clears the prompt, then Up on the empty prompt restores it.

This is particularly relevant when the prompt includes more than typed instructions. Pasted material and images are explicitly included in the recovery behavior, so the feature covers the contents of that cleared draft.

Set the number of review findings

The new /code-review --max-findings <n>|all option controls how many findings a review reports. You can choose a number, request all findings, or restore the default:

/code-review --max-findings 5
/code-review --max-findings all
/code-review --max-findings default

The setting persists until you reset it with default. Treat it as a continuing preference when moving between reviews.

The release does not specify the numeric default, so there is no verified default count to quote. If you want a known reporting limit, set an explicit number; if you want to return to the built-in behavior, use the reset command. The CLI reference is a useful companion for broader command-line workflows.

Plugin hooks gain clearer agent tracking

Claude Code 2.1.289 adds plugin-hook information for teammate creation and agent state. The relevant changes are agent.spawn for teammates, one agent ID across plugin-hook events, and idle or waiting states in $.agent.list().

For plugin authors, these additions provide more information about agent activity. A consistent ID gives events a common agent identifier, while idle and waiting states add detail to the list returned by $.agent.list(). The brief does not supply a complete implementation example, so these additions should be treated as API capabilities rather than a ready-made orchestration recipe.

Mod loading and UI fixes

The same release fixes installed mods failing to load in the first session after an upgrade. It also addresses stale local-marketplace plugin copies and several mod UI failures.

Those fixes are relevant when validating an extension after updating Claude Code. Include the first session after an upgrade in your checks, and confirm that the installed plugin reflects the expected local-marketplace version.

Version 2.1.288 had already added $.ui.selection(). It returns the last fullscreen text selection and, when that selection is contained within a single transcript row, the row itself. That gives mod authors selection information with a documented condition for receiving the associated row.

Permission handling on managed machines

Version 2.1.289 also fixes permission behavior involving user-installed mods. On organization-managed machines, deny and ask rules on nested parts of compound shell commands are preserved against mod approvals.

Other fixes address Read deny rules through IDE symlinks and Bash deny or ask rules under sandbox auto-allow. These changes deserve attention alongside the new mod capabilities: extension behavior and permission enforcement are both part of evaluating an upgrade.

Session navigation, MCP, and gateway compatibility

Several smaller Claude Code new features affect navigation and integration work.

In 2.1.287, the agents view gained an n:<text> filter that matches session names and tasks. Version 2.1.288 then added Ctrl+F to search sessions by name and Alt+↑/↓ to jump between groups in the agents view. Those shortcuts and rename can be rebound in keybindings.json.

For restored sessions, 2.1.288 includes several --resume fixes that preserve restored context, final responses, and earlier thinking. These are reliability changes to resumed work, separate from the recovery of a prompt cleared with Ctrl+C.

MCP URL prompts

Version 2.1.287 added URL prompts for MCP servers using protocol 2025-11-25. If a server stops connecting, the release gives "bareElicitationCapability": true as a compatibility setting.

That setting is a troubleshooting detail tied to the release’s MCP change. Check it in the context of the affected server rather than treating it as a universal configuration requirement. The MCP guide offers a starting point for the surrounding integration workflow.

Structured outputs

Version 2.1.288 introduced CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS, which disables structured outputs for environments such as gateways that reject them.

The brief provides the environment variable’s purpose but does not give a specific assignment value or shell command. If structured-output rejection affects your environment, this is the configuration option to investigate; the release does not establish that every gateway needs it.

Usage limits to keep in view

The October feature changes sit alongside a September adjustment to subscription usage limits. Since September 14, weekly limits have been permanently 25% above pre-promotion levels for Pro, Max, Team, and seat-based Enterprise.

The earlier temporary 50% increase ran from May 13 through September 13, 2026, ending at 11:59 p.m. PT. Free and consumption-based Enterprise seats were excluded from that promotion.

Keep those percentages separate when checking your current allowance. The permanent increase and the expired promotion use different figures; the October releases do not establish a return to the promotional level.

Use /usage to view your limits. For more context on the relevant subscription constraints, see the Claude Code rate-limits guide.

Key takeaways

  • Claude Code 2.1.287 introduced TypeScript mods that ship inside plugins and can modify prompts, UI, built-in features, and tool calls in the CLI and desktop app.
  • Mods are not sandboxed. They run with Claude Code’s machine access, and the announcement advises installing them only from trusted sources.
  • Version 2.1.288 restores cleared drafts: after Ctrl+C clears a prompt, press Up on the empty prompt to recover it, including pasted text and images.
  • Review findings are configurable and persistent. Use /code-review --max-findings 5, all, or default; the numeric default is not specified.
  • Version 2.1.289 expands agent-related plugin hooks and fixes mod loading, UI, and permission handling. Use /usage to check your current limits separately from these feature changes.

FAQ

What are Claude Code mods?

Claude Code mods are small TypeScript functions packaged inside plugins that can rewrite prompts, add UI, replace built-in features, and block, rewrite, or retry tool calls. They work in the CLI and desktop app, run with Claude Code’s machine access, and are not sandboxed.

How do I create and test a TypeScript mod?

The 2.1.287 snapshot documents an exported register(on, options) function, handlers shaped ($, e, next), and a plugin containing .claude-plugin/plugin.json, hooks/hooks.json, and TypeScript under hooks/. Use /plugin-types to write declarations; the documented example commands are claude --plugin-dir mods/diff and claude plugin test mods/diff.

How do I recover a prompt accidentally cleared with Ctrl+C?

In Claude Code 2.1.288, press Up on the empty prompt after Ctrl+C clears it. This restores the cleared draft, including pasted text and images.

How do I change or reset the code-review findings limit?

Use /code-review --max-findings 5 for a numeric limit or /code-review --max-findings all to request all findings. The choice persists until /code-review --max-findings default, and the release does not specify the numeric default.

A

AI Prime Tech publishes this blog. Articles are drafted with AI assistance; check version-specific details such as model IDs, prices and limits against the official documentation before relying on them.

Get cheaper Claude API access

One API key for Claude Opus 5.5, Sonnet 5, Haiku 4.5 and Fable 5.1, plus GPT-6 models. Pay as you go, no subscription.

Get Your API Key →
AI Prime Tech is an independent third-party API gateway. Claude™ and Anthropic® are trademarks of Anthropic, PBC. No affiliation or endorsement is implied.