Summary
Member of Technical Staff at Anthropic, working on Claude Code. Authored the session management and 1M context guide.
Key Points
- Authored practical guide on Claude Code session management, context compaction, and context rot
- Key insights:
- Context rot: model performance degrades as context grows
- Rewind often better than correcting: keep useful file reads, drop failed attempt
- Subagents get fresh context: only final report comes back
- 1M context enables longer tasks but requires proactive compaction
Evidence Timeline
- 2026-04-17: Created from “Using Claude Code: session management and 1M context” article
相关页面
birgitta-bockeler, chaoyue-he, chelsea-finn, jinse-chuanshuo-dacongrming, martin-kleppmann, omar-khattab, scale-ai