Claude Code v2.1.293: Haiku 5.5, subagent status-line agentType, deferred-tool control for mods#

Published by

Part of the Claude Code Version Tracker series. | Official Env Vars | Official Changelog

Claude Code v2.1.293 adds Claude Haiku 5.5 (claude-haiku-5-5), a 1M-context model priced at $0.10 input and $0.50 output per million tokens and set as the default Haiku on the Anthropic API.[1] It also stops Claude from retracting or redoing finished work after a context compaction and fixes a memory leak where an HTTP MCP connection held every request until it closed. It follows v2.1.292.

Haiku 5.5 joins the model lineup#

The new model is claude-haiku-5-5, shown as "Haiku 5.5". It carries a 1M-token context window. Input costs $0.10 and output $0.50 per million tokens; for prompts over 100K tokens the rate rises to $0.50 input and $2.50 output.[1] The changelog makes it the default Haiku on the Anthropic API.

New control for mods and subagents#

Two scripting additions land alongside the model. A custom status-line script configured for subagents now receives agentType in its JSON payload, so one script can tell custom subagent types apart.[1]

For mods, $.tool.register takes isDeferred. Set to false, it lists the registered tool's schema in the prompt from the start instead of behind tool search.[1] The deferred-tool path it controls is the same one that keeps most tool schemas out of the prompt until a search surfaces them.

Fixes across compaction, backgrounding, and MCP#

Beyond the compaction and HTTP MCP fixes in the opening, a message sent while Claude was working could be lost when the left arrow moved the session to the background; now, if a queued message cannot move with the session, the left arrow stays put and says so.[1] The /model effort selector no longer wraps past the highest or lowest level, which could accidentally save Low as a model's default effort. /tui no longer disconnects Claude in Chrome in a session started with --chrome, and no longer ignores --no-chrome.

Claude is no longer told to continue or message subagents with SendMessage in sessions where a host, a permission rule, or a --tools list has removed that tool, and subagents are no longer told a built-in tool is disabled for the whole session when only their own tool list leaves it out.[1] Startup for Team and Enterprise organizations fetches policy and managed settings earlier and retries a stalled request after three seconds.

What These Tell Us#

The model lineup now reaches a 1M-context Haiku at the low end of the pricing table.

The scripting surface for mods and subagents grew in two concrete places. A subagent status-line script can branch on agentType, and a mod can choose whether a registered tool's schema sits in the prompt from the start or waits behind tool search. Both give a script author a decision the build previously made for them.

The rest of the release bounds failures around context compaction, session backgrounding, and MCP connections, where a lost queued message, a leaked set of HTTP requests, or finished work being redone each left a visible mark on a session.

Sources#

  1. Claude Code Official Changelog, v2.1.293 release notes

This analysis is conducted for educational and research purposes under fair use principles. All trademarks and software referenced belong to their respective owners.

Let's Connect

BlueskyLinkedInThreads

Previous Claude Code analysis: v2.1.292 (October 6, 2026)

All 67 Claude Code analyses: Claude Code Version Tracker · Official Claude Code changelog

← Back to all posts