JA EN
shiichan

Claude Code v2.1.299: Idle Compaction Keeps Your Next Message Light

Claude Code Release ★★★★ View Markdown
#anthropic#ai#devops

Hi, it's Shii-chan! Claude Code v2.1.299 is out, so here are the parts I think matter most.

Claude Code Release github.com

What was announced?

The Claude Code Release page on GitHub published the notes for v2.1.299. There are many entries under Added, Changed, Fixed, and more, so I picked the ones most likely to touch everyday use. Check the original release notes for the full list.

Why it matters

The longer a Claude Code conversation runs, the more the prompt cache expiry matters. This release adds a feature aimed squarely at the slow next message you get after the cache runs out.

It also includes many changes that prevent unintended behavior, such as stricter MCP allow rules and skill name handling.

What changes

First, the headline additions:

  • Idle compaction in the terminal: a conversation of 200K+ tokens left idle for about 54 minutes is compacted before its one-hour prompt cache expires, so your next message doesn't reprocess everything
  • CLAUDE_CODE_DISABLE_FILE_MENTIONS environment variable: for @ file mentions, Claude gets the paths and reads the files with its own file tools, instead of getting the contents
  • /effort and /fast in agent view set what new sessions start with; a bare /model, /effort, or /fast there opens a picker
  • --shutdown-drain-sec for claude self-hosted-runner orchestrator: on SIGTERM it stops polling and gives the poll in flight and running spawn hooks up to that many seconds to finish

Dive Deep

Configuring idle compaction

Change the minimum with CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS (the default is 200K). To turn it off, set this in settings.json. Both need v2.1.290 or later.

{
  "idleCompaction": false
}

New things for mods

There are several additions for plugin mods too:

  • A prompt.submit hook of a mod installed by managed settings can read the bytes of an image or document attached to the prompt via $.prompt.media
  • Mods installed by managed settings can call $.session.identity() to see how the session is signed in, and its organization and email when managed forceLoginOrgUUID lists that organization
  • A mod's Text gained size, weight, mono, and tabular, and its Box gained named spacing steps from xs to xl. bold is deprecated in favor of weight
  • A mod's skill.prompt hook now receives the skill's frontmatter, source, plugin name, and file path

Changes worth knowing about

There were 25 entries under Changed. These are the ones that alter behavior:

  • MCP allow rules reach only the server they name. mcp__a no longer allows a server named a__b or a_; write mcp__{server}__* with the server's whole normalized name
  • Agent tools: lists likewise take only the named MCP server's tools. List a server whose name contains __ as mcp__{whole name}__*
  • Requests to claude.ai's connector proxy and to Anthropic-hosted MCP servers no longer follow HTTP redirects
  • The resume command printed at exit names the session by its ID rather than its name, so nothing in a session name can run when the line is pasted
  • UserPromptSubmit hooks run before a skill or custom slash command expands, so a hook that blocks the prompt also stops its embedded shell commands
  • MCP tool descriptions and server instructions from Anthropic's first-party MCP servers at api.anthropic.com reach Claude in full instead of being cut at 4,096 characters
  • A skill or command whose file or folder name has hidden, direction-changing, or blank-like characters, <, or > is no longer offered until renamed
  • The errors shown when /ultrareview can't upload a repository no longer suggest git add ., which could stage untracked files such as .env

Wrap-up

  • Conversations of 200K+ tokens are compacted after about 54 idle minutes, before the cache expires (disable with idleCompaction)
  • CLAUDE_CODE_DISABLE_FILE_MENTIONS makes @ mentions pass paths only
  • MCP allow rules and agent tools: lists now match server names strictly
  • Resume commands, hook ordering, and skill name validation got safety tweaks

This one is for people who leave long sessions sitting, teams writing MCP allow rules, and anyone building mods!

Related posts