[Hackathon] fix(prompt): a follow-up prompt on an already-ended stdin never resolves — setup exits 13 as 'unsettled top-level await' #639

Verified against (commit SHA or released version)

commit a18d37c (main HEAD)

Discord Username / User ID

sati1325

What does this improvement do?

When the CLI shows a second prompt on an input stream that has already ended — e.g. the setup target-confirmation loop (CONFIRM_TARGETS_MAX_ATTEMPTS re-prompt after an invalid answer) or agent install\'s target prompt — promptText/promptSecretnever resolves.readLineonly callsisInputEnded(input) when leftover buffered input exists (src/lib/prompt.tslines 138-147), then unconditionally attachesinput.on('end', …)at line 150. Butendalready fired, so the promise never settles. With nothing left keeping the event loop alive, Node aborts withWarning: Detected unsettled top-level await … program.parseAsync` and exits with code 13; if another handle keeps the loop alive, the prompt simply hangs forever.

Repro (real PTY): run testsprite setup (no flags) and at Install skills for … [claude]: type a typo like clade and press Ctrl-D Ctrl-D (terminal EOF). Actual: after unknown target \"clade\" the re-prompt prints, then the process dies with the unsettled-TLA warning and status 13 — no error envelope, no telemetry flush. Expected: the re-prompt notices stdin has ended and resolves '' (accepting the default, like the first prompt's onEnd does) or rejects with a proper error — exiting with a documented error code, not Node's internal unsettled-await one.

Same failure at unit level on a18d37c: const input = new PassThrough(); input.end('typo\n'); await promptText('q1', {input}); await promptText('q2', {input}) — the second call never settles (verified with a Promise.race timeout).

Real-world impact: exit 13 is documented in DOCUMENTATION.md as "Feature gated (plan entitlement or per-account flag)", so scripts and CI wrappers misread this crash as a billing/entitlement failure. A typing mistake + Ctrl-D — a normal terminal habit — silently kills interactive setup.

Details / implementation notes

Root cause: readLine (src/lib/prompt.ts) checks isInputEnded(input) only inside the pending !== undefined branch (lines 138-147). With no buffered leftovers it falls through and attaches on('end', onEnd) to a stream whose readableEnded is already true — the event already fired and will never fire again, so the promise stays pending. Once stdin's EOF stops holding the event loop open, Node's unsettled-top-level-await path exits with code 13 and prints the internal warning.

Fix: after processing any pending text, check isInputEnded(input) (or readableEnded) unconditionally before attaching listeners, and finish() when ended — resolving with whatever was typed ('' if nothing), matching the first prompt's onEnd semantics. Optionally track ended streams (e.g. a WeakSet marked in onEnd) so later prompts on the same stream resolve immediately. Regression test: input.end('typo\n'); await promptText(); await promptText() must resolve '' rather than hang. The same helper backs agent install's target prompt, so the fix covers both interactive paths.

A PR with the fix + regression test will follow referencing this issue.

Confirmations

• I have searched existing issues and this is not a duplicate.
• I have provided my Discord identity above for reward coordination.