Skip to content

Claude Code

Claude Code 2.1.199: improved sub-agent and background task stability

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: Claude Code versions and changes →

Key fixes focus on error handling. Streaming responses are no longer discarded if the API returns an overloaded status mid-output: partial results are now saved with an incompleteness flag. Sub-agents interrupted by limits or server errors now return partial work to the parent instead of failing silently.

Sub-agent API errors are no longer reported as successful results. The background agent daemon on Linux no longer terminates itself and all active agents every 50 seconds following an incorrect shutdown. The cold start issue on macOS via SSH, which triggered a Could not switch to audit session error, has been resolved after appearing in version 2.1.196.

Transient 429 errors that occur without exceeding actual limits now trigger a backoff for subscribers. The CLAUDE_CODE_RETRY_WATCHDOG variable now raises the retry counter to 300 and removes the previous cap of 15 for CLAUDE_CODE_MAX_RETRIES.

Community projects

claude-real-video: Local video processing for any LLM

A 10-minute static slide is 600 near-identical frames. claude-real-video compresses them to one, while capturing fast cuts frame-by-frame via scene detection instead of timers. Everything runs locally: yt-dlp pulls the video, Whisper transcribes, and the output is a folder of frames and a MANIFEST.txt that any LLM can ingest.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.