[liususan091219]

What the owner saw

A terminal offering three options, and a "needs your attention" card offering two — the first missing, the rest renumbered.

Two independent causes

Both measured against the real Claude Code permission dialog, not a hypothetical:

1. A wrapped option label split the block. option_blocks() ended a block at any non-option line, so a hard-wrapped label broke the run in two. parse_options() then returned only the trailing block, dropping every option before the wrap:

  1. Yes, and allow this command for the rest of
     this session          <- ends the block here
  2. Yes
  3. No, and tell Claude what to do differently

→ before: ['Yes', 'No, and tell Claude what to do differently'] · after: all three, label rejoined.

2. Any yes+no dialog collapsed to two buttons. requirement_for() rendered No/Yes whenever it found a yes and a no, regardless of what else was offered. Claude Code's actual permission dialog has three — Yes, Yes, and don't ask again…, No — so the middle option had no button, and no keystroke could reach it.

A dialog that genuinely offers two still renders No/Yes. Three or more now render one button per option in the terminal's own order, and keys_for() walks the caret to the clicked row.

Reviewer note — this reverses a prior deliberate choice

test_a_permission_dialog_is_a_no_yes_card_in_the_owners_words pinned the No/Yes shape for the 3-option dialog, by earlier owner preference. V36/V42 (owner, 2026-11-06) supersede it for 3+ options. I replaced that test rather than deleting it, and added a new one pinning the two-option case it was protecting, so the old intent is still covered where it still applies.

Regression evidence

Regression test: tests/hitl-tui-gate.test.py

Before fix: FAIL — reverting the collapse guard alone leaves 8 failures; reverting the wrap fix alone leaves 3. Run: python3 tests/hitl-tui-gate.test.py

After fix: PASS — 37/37. Same command.

Each fix was reverted independently to confirm both are load-bearing rather than one masking the other.

Not in this PR

V36 (card copy) and V44 (one status after resolve) are separate — V44 overlaps the web-side card owned by another agent and is waiting on that coordination.

🤖 Generated with Claude Code

https://claude.ai/code/session_01ByiYdfob1ZYnq5PHYxo8qb

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Susan Xueqing Liu seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[github-actions]

@cla-assistant check

[github-actions]

Coverage Gate

✅ Diff coverage PASSES the 95% bar. Whole-tree (informational): 93%.

Diff Coverage

Diff: origin/main...HEAD, staged and unstaged changes

• src/hitl/tui_gate.py (100%)

Summary

• Total: 8 lines
• Missing: 0 lines
• Coverage: 100%