[DeadWaveWave]

💡 Change Scope

• Small Change: Fast feedback, localized UI/logic, low-risk.
• Large Change: New feature, cross-boundary logic, runtime-risk (persistence, IPC, lifecycle, recovery).

📝 What Does This PR Do?

Fixes stale terminal replies being typed into the shell after a terminal's xterm instance is recreated.

When OpenCove recreates a terminal's xterm instance (switching the terminal theme mode, restoring a terminal, and — with the upcoming transparent terminals — switching the terminal background), it replays the terminal's history into the new instance. If that history contains queries a program sent earlier — OSC colour queries (ESC ] 11 ; ?), cursor position reports (ESC [ 6n), device attributes (ESC [ c) — the new instance answers them again, and the answer arrives in the PTY as input to whatever runs now. Found while writing the transparent-terminal E2E: after a recreation, zsh received ESC ] 11 ; rgb:0909/0e0e/1212 BEL and printed command not found: 11.

• stripTerminalQuerySequences (src/shared/terminal/automaticTerminalSequences.ts) removes the CSI queries the input bridge already recognises plus OSC 4 / 10–19 colour queries, and keeps everything else (colour sets, titles, cursor movement, styles).
• hydrateFromSnapshot strips replayed snapshots before writing them to xterm. Live output that arrives during hydration keeps its existing routing, so programs that ask while a terminal hydrates still get answers.

---

🏗️ Large Change Spec (Required if "Large Change" is checked)

Not a Large change: one pure function and one call site in the replay path.

Invariant: replayed history never produces input to the PTY; live queries are still answered.

---

✅ Delivery & Compliance Checklist

• The local quick gate passes: pnpm verify:quick / pnpm pre-commit is green; any additional focused tests are reported above.
• Full CI passes on the final revision before merge; local quick checks are not reported as full regression.
• I have signed the CLA if required (see CLA.md).
• I have included new tests to lock down the behavior (or explicitly stated why it's untestable).
• I have strictly adhered to the DEVELOPMENT.md architectural boundaries.
• I have attached a screenshot or screen recording (if this touches the UI).
• I have updated the documentation accordingly (if adding a feature or changing a contract).

Unit: automaticTerminalSequences.spec.ts (each query kind removed; colour sets, titles, cursor movement and styles kept). The end-to-end proof is the transparent-terminal E2E in the follow-up PR, which recreates xterm after an OSC 11 query and checks the next command runs cleanly (it failed before this fix).

Local (macOS arm64): pnpm verify:quick passed with this change (related Vitest 330 files / 1563 tests, native contracts 6 / 9, representative E2E 18 tests, run together with the follow-up); on this branch alone the hydration and sequence unit tests (28 files / 178 tests) and the terminal, recovery and reattach E2E files (45 files: 65 passed, 6 skipped, 1 failed). The failure was workspace-canvas.terminal-agent-restart.spec.ts:39 (codex): the restored sentinel appeared and then the screen was redrawn blank. Instrumenting the stripper showed it removes nothing in that test, so the replayed data is identical to main; across repeats it failed 2 of 13 times on this branch and passed 12 of 12 on main, which reads as a rare timing flake that this change does not touch. CI will show if it recurs.

Full CI on the final revision 8bf03a56: all 11 checks passed (run 38828073866, attempt 2). Attempt 1 failed only in web-canvas-continuity-ubuntu: workerWebCanvas.terminal-live-reattach-geometry.spec.ts:10 could not seed its state (sync.writeState failed: persistence.invalid_state), before any terminal output was replayed; it passed on the rerun. workspace-canvas.terminal-agent-restart passed on all three macOS shards.

📸 Screenshots / Visual Evidence

No UI change.

🤖 Generated with Claude Code