I hit this while pasting ~8K of DOM markup into Claude Code and iterating on selectors for a Chrome extension. About 40 minutes in, compaction fired and the summary said "user provided DOM markup" but the actual content was gone. Claude started guessing at selectors it had seen 20 minutes earlier.

The transcript with the original markup is still sitting at ~/.claude/projects/ as a .jsonl file — the compaction summary just has no pointer back to it.

I found 8+ open issues on the repo describing different symptoms of this same root cause. The proposal is to add line-range annotations to compaction summaries so Claude can surgically recover just the chunk it needs from the transcript on demand. Zero standing token overhead.

Curious if others have hit this in different scenarios or found workarounds that actually stick.

This is a real pain point. When a long conversation hits the context limit, compaction summarizes everything to fit — but the summary inevitably loses nuance. You end up with the agent "forgetting" file paths it already explored, decisions it already made, or code it already wrote.

The underlying issue is that there is no reliable way to know what will be important later. A file path mentioned in passing early on might be critical for the final step. Compaction is lossy compression on a signal whose importance distribution you do not know ahead of time.

Practical workaround: persist critical state to disk files as you go. Write decisions, file paths, and intermediate results to a "memory" file that survives compaction. The agent can re-read it after compaction restores context. It is clunky but it works.