fix(web): suppress xterm.js terminal query auto-responses

[konard]

Summary

Fixes #271 — the web terminal was showing raw escape sequences such as ^[]10;rgb:f4f4/f7f7/fbfb^[\ and ^[[?1;2c inside Claude Code's prompt, breaking navigation and rendering.

Root cause

xterm.js@5.3.0 answers TUI capability probes out of the box:

• OSC 4/10/11/12 color queries (\x1b]10;?\x1b\\, etc.)
• CSI ... c device attributes — DA1, DA2, DA3
• CSI ... n device status / cursor position reports

The replies are emitted via Terminal.onData, which the web frontend forwards to the host PTY as if the user typed them. Claude Code (and other TUIs) read those bytes as keystrokes inside their prompt loop and render them verbatim — exactly what the screenshot in the issue shows.

Fix

A new module (packages/app/src/web/terminal-query-suppression.ts) installs custom OSC/CSI handlers immediately after the Terminal is constructed in createTerminalRuntime:

• For OSC 4/10/11/12 the handler returns true only when the payload contains a ? segment (a query). Plain set-color payloads return false so the default theme-change handler still applies.
• For CSI ... c (DA1/DA2/DA3) and CSI ... n (DSR/CPR) the handler always returns true so xterm never reports back to the PTY.

Handlers are tracked as disposables, so callers (and the unit tests) can roll the registration back without touching xterm's internal parser state.

Test plan

• bun run --bun vitest run tests/docker-git/terminal-query-suppression.test.ts — 7 new unit tests covering query detection, handler registration order, OSC fall-through for set commands, CSI consumption, and disposal cleanup.
• Full test suite passes.
• bun run typecheck passes.
• Linter passes.
• bun run build:web succeeds.
• Manual: start the web build, attach a TUI inside the container, run printf "\\033]10;?\\033\\\\" — no echoed garbage in the prompt.

Files

• packages/app/src/web/terminal-query-suppression.ts (new) — suppression module
• packages/app/src/web/terminal-panel-runtime-core.ts — wire suppression into createTerminalRuntime
• packages/app/tests/docker-git/terminal-query-suppression.test.ts (new) — unit tests
• experiments/terminal-query-suppression.md — root-cause notes
• experiments/verify-suppression.html — manual browser verification scaffold

Fixes #271

[konard]

Working session summary

CI is running. Done — committed the fix, pushed, updated PR #272's title/body, and marked it ready for review.

PR: #272

Summary of changes shipped on issue-271-6e1a0e1d8cfa:

• packages/app/src/web/terminal-query-suppression.ts — new module that overrides xterm.js OSC 4/10/11/12 and CSI DA1/DA2/DA3/DSR/CPR handlers so terminal probes are consumed without replying through onData.
• packages/app/src/web/terminal-panel-runtime-core.ts — wires the suppression into createTerminalRuntime right after the Terminal is constructed.
• packages/app/tests/docker-git/terminal-query-suppression.test.ts — 7 unit tests covering query detection, handler registration, fall-through for set-color commands, and disposal.
• experiments/terminal-query-suppression.md + experiments/verify-suppression.html — root-cause notes and manual verification scaffold.

CI is currently pending across all jobs.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $5.497444

📊 Context and tokens usage:

Claude Opus 4.7: (2 sub-sessions)

1. 117.4K / 1M (12%) input tokens, 27.2K / 128K (21%) output tokens
2. 39.1K / 1M (4%) input tokens, 3.3K / 128K (3%) output tokens

Total: (2.0K new + 134.5K cache writes + 7.3M cache reads) input tokens, 35.7K output tokens, $5.405814 cost

Claude Haiku 4.5:

• 40.9K / 200K (20%) input tokens, 3.3K / 64K (5%) output tokens

Total: (593 new + 40.3K cache writes + 241.6K cache reads) input tokens, 3.3K output tokens, $0.091631 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Main model: Claude Opus 4.7 (claude-opus-4-7)
• Additional models:
  • Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (2959KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🔄 Auto-restart triggered (iteration 1)

Reason: CI failures detected

Starting new session to address the issues.

---

Auto-restart-until-mergeable mode is active. This run will stop after 5 restart iterations.

[konard]

Working session summary

All 15 PR checks are SUCCESS — including the previously failing Lint Effect-TS.

Summary

The Lint Effect-TS check was failing because the new test file packages/app/tests/docker-git/terminal-query-suppression.test.ts had five as never casts (lines 74, 80, 92, 101, 109), and the lint rule no-restricted-syntax forbids TSAsExpression/TSTypeAssertion outside of src/core/axioms.ts.

Fix (commit a00ffc3): Replaced the Terminal-typed parameter on installTerminalQuerySuppression with a structural type TerminalQuerySuppressionTarget that only declares the parser.registerOscHandler/parser.registerCsiHandler methods the function actually uses. The real xterm Terminal already satisfies this structural shape (parameter contravariance + return covariance), and the test's mock now satisfies it too — so no casts are needed.

Verified locally: bun run --cwd packages/app lint:effect clean, full app tests 262/262 pass. PR 272 is now green on all 15 checks.

PR: #272

---

This summary was automatically extracted from the AI working session output.

[konard]

🔄 Auto-restart-until-mergeable Log (iteration 1)

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $2.200154

📊 Context and tokens usage:

• 84.4K / 1M (8%) input tokens, 12.6K / 128K (10%) output tokens

Total: (52 new + 78.3K cache writes + 2.8M cache reads) input tokens, 12.6K output tokens, $2.200154 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Model: Claude Opus 4.7 (claude-opus-4-7)

📎 Log file uploaded as Gist (4076KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🔄 Auto-restart triggered (iteration 2)

Reason: Uncommitted changes detected

Starting new session to address the issues.

---

Auto-restart-until-mergeable mode is active. This run will stop after 5 restart iterations.

[konard]

Working session summary

Everything is in order:

• Working tree: clean (removed stale ci-logs/ artifacts)
• Branch: up to date with origin and merged with latest upstream/main (no conflicts)
• PR fix(web): suppress xterm.js terminal query auto-responses #272: OPEN, MERGEABLE, mergeStateStatus CLEAN, not draft
• CI: all 15 checks passing (Build, Lint, Lint Effect-TS, Test, Types, Snapshot, 7 E2E jobs, Dist deps prune × 2)
• Local tests: 7/7 terminal-query-suppression tests pass; effect-ts lint clean

PR: #272

---

This summary was automatically extracted from the AI working session output.

[konard]

🔄 Auto-restart-until-mergeable Log (iteration 2)

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $0.828325

📊 Context and tokens usage:

• 48.5K / 1M (5%) input tokens, 4.3K / 128K (3%) output tokens

Total: (34 new + 52.1K cache writes + 787.8K cache reads) input tokens, 4.3K output tokens, $0.828325 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Model: Claude Opus 4.7 (claude-opus-4-7)

📎 Log file uploaded as Gist (4639KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag