Developer Productivity with Claude
Your team uses the Claude Code SDK to run long-running developer sessions that span dozens of tool calls — reading files, running tests, and iterating on a complex refactor. Developers report that after automatic compaction triggers mid-session, Claude loses track of the acceptance criteria defined at the start of the session and begins re-introducing patterns it was explicitly told to avoid. Logs confirm the compact_boundary system message is emitting correctly, and the compaction itself is working as intended. What is the most effective fix?