[bug] bug unable to type anything after starting claudish

[dev-seahouse]

Please see the video.

Screencast_20260323_213410.mp4

This seem to happen since v6.0.0, when i installed v5+ it was working.

It looks like something wrong with stdin handling.

@erudenko

[claudish-bot]

@dev-seahouse This looks like a stdin handling regression. In v5.4.0 there was explicit stdin cleanup code (pause + removeAllListeners) that ran after interactive prompts to prevent watchers from stealing stdin before Claude Code inherited it — that code got removed during the v5.19.0 PTY refactor.

What's likely happening:

• Interactive prompts (API key setup, telemetry consent in telemetry.ts:565) attach readline listeners to process.stdin
• Without cleanup, those listeners stay active and consume stdin data
• Claude Code spawns with stdio: "inherit" (claude-runner.ts:430) but stdin is already drained/locked

Workaround to test: Try running with --no-telemetry or after you've already consented to telemetry (so no prompts fire). If typing works then, that confirms the diagnosis.

Can you also confirm whether you're running with Bun or Node? The PTY path (pty-diag-runner.ts) handles stdin differently than the standard stdio: "inherit" path, so knowing which codepath you're hitting would help narrow this down.

[dev-seahouse]

im using bun

[claudish-bot]

@dev-seahouse Thanks — Bun means you're hitting the standard stdio: "inherit" path in claude-runner.ts, not the PTY codepath. That's consistent with the stdin-stealing theory.

Could you try the --no-telemetry flag workaround I mentioned? If typing works with that, it confirms the readline cleanup regression and makes the fix straightforward. Also, does this happen every launch or only on first run (when consent prompts fire)?

[dev-seahouse]

same when running under node 24, looks like not a bun issue

[claudish-bot]

@dev-seahouse Good data — rules out a Bun-specific issue. Since it reproduces on both runtimes, this is definitely in claudish's stdin handling, not the JS engine.

Could you still try --no-telemetry to test whether the interactive prompts are the culprit? If it's broken even with that flag, the regression is in the spawn logic itself (claude-runner.ts) rather than readline cleanup — different fix path.

Also, does the screenshot show a fresh install (first-run prompts) or a subsequent launch where you'd already consented to telemetry?