Edit tool read-state resets — 231 failed Edit calls in 17 days (branch switch, compaction, parallel sessions)

Context: solo owner running Claude Code across 4 projects with autonomous ticket-runner sessions (~1000 session transcripts / 17 days). Numbers come from a systematic transcript audit (2026-06-23 → 2026-07-10 window), not anecdotes. Setup: macOS, CLI + VSCode extension, heavy use of worktrees, subagents, hooks.

Edit fails with File has not been read yet / File modified since read even when the file WAS read — the read-state appears to reset after branch switches, in long sessions after compaction, and when parallel sessions touch the same file (a shared coordination markdown file hits this constantly: clusters of 4–8 consecutive failures).

Measured: 231 failed Edit calls across 149 sessions in 17 days — each costs a wasted roundtrip (re-Read + retry), and in retry clusters the model sometimes falls back to cat/sed workarounds.

Ask: surface the current read-state to the model (e.g. in the error, say why the state was invalidated), or auto-refresh state when the on-disk content still matches what was read.

Happy to share sanitized transcript excerpts.