# Claude Code v2.1.217 Adds a Runaway-Subagent Guard and Fixes a Memory Leak!

Hey, it's Shiichan again! There's a new Claude Code release, so let's dive right in.

## What was announced?

Claude Code Release published the latest version of the Claude Code CLI, v2.1.217. Alongside two new features, it packs in a dense set of subagent safety guards plus performance and security fixes.

## The story so far

Until now, Claude Code had a handful of quiet-but-real issues.

- Truncated MCP tool outputs still kept the full untruncated result in memory for the rest of the session, causing a memory leak
- A failed Windows auto-update could leave `claude.exe` missing entirely
- Background session isolation didn't canonicalize symlinked working directories, which could let sessions escape their workspace folder
- Auto-compact never triggered for Claude Opus 4.8 on Bedrock, and `/compact` itself failed once you were over the limit
- Corporate mTLS, TLS-verify, OAuth scope, and proxy settings were ignored in Claude Desktop sessions
- Subagents could spawn nested subagents, letting a single message fan out an unbounded number of them
- `--max-budget-usd` didn't actually stop background subagents once the cap was reached

## What changes

This release fixes all of that, and the subagent safety guards stand out the most.

- Concurrently-running subagents are now capped (default 20), configurable via `CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS`
- Subagents no longer spawn nested subagents by default; set `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH` to allow deeper nesting
- Once `--max-budget-usd` is reached, new subagent spawns are denied and running background subagents are halted

Together, these cut the risk of a single message fanning out into unbounded subagents and runaway cost.

The other fixes landed too: the MCP tool output memory leak, the Windows auto-update `claude.exe` disappearance, the symlink-based workspace escape, the Opus 4.8 on Bedrock auto-compact issue, and Desktop ignoring corporate network settings.

On the feature side, the prompt input now supports emoji shortcode autocomplete — type `:heart:` to insert ❤️, or `:hea` to see suggestions (disable it with the `emojiCompletionEnabled` setting). There's also a new warning when transcript writes fail (like when disk is full) or session saving is off due to an inherited environment variable, instead of silently losing your history.

## Dive Deep

A bunch of smaller fixes landed too.

- The login-expiry warning now appears 3 days before expiry instead of 5
- The frontend-design plugin suggestion is now capped at 3 lifetime impressions instead of repeating indefinitely
- Footer PR badge links are now clickable even when hyperlink support can't be detected, like over ssh or tmux (opt out with `FORCE_HYPERLINK=0`)
- Fixed `CLAUDE.md`/`SKILL.md` `paths` frontmatter with many brace groups OOM-killing or stalling the CLI at startup — brace expansion is now budget-bounded
- Fixed `--resume`/`--continue` and `/resume` failing with a TypeError when a transcript has a malformed attachment entry
- Fixed Remote Control sessions not showing a pending permission prompt or dialog to viewers who connected after it appeared
- Fixed background shells sometimes becoming impossible to stop after sending a session to the background (`/background` or `←`) or on exit on a heavily loaded machine, most visible on Windows

## Wrap-up

- Added safety guards against subagent runaway and cost blowups: a concurrency cap, no nested spawning by default, and an enforced `--max-budget-usd`
- Fixed a memory leak from MCP tool output, a Windows auto-update bug that could delete `claude.exe`, and a symlink-based workspace escape
- Added emoji shortcode autocomplete in the prompt input and a warning when transcript writes fail
- Plus smaller fixes: login-expiry warning timing, footer links, and a `/resume` TypeError
- A must-know update for anyone leaning heavily on subagents or background sessions, for both safety and cost control!