Prompt suggestions stopped appearing entirely (v2.1.207, Windows 11) — worked reliably until 2026-07-12, dead since, across restarts and with all flags explicitly enabled

Description

Prompt suggestions (the greyed-out predicted next-message text accepted with Tab+Enter) stopped appearing entirely on the night of 2026-07-12 and have not appeared once since, across many sessions over two days.

Before that they fired reliably and frequently for weeks on this machine — most consistently after running a project slash-command skill whose output ends with a recommendation (the suggestion would mirror the recommendation). The same workflows now produce zero suggestions.

What makes this report different from the closed ones

#31077 and #30180 were closed as needs-repro / stale. This report has a clean before/after boundary and config verification:

• Worked reliably until the night of 2026-07-12, zero occurrences since — a sharp cutoff, not a gradual or intermittent failure. Timing coincides with an auto-update (running 2.1.207, released around then).
• Config verified, not assumed. When it stopped, nothing was set: no CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION env var anywhere, no promptSuggestionEnabled key — i.e. the feature was running on defaults (which per the settings schema default to enabled).
• Explicitly enabling changed nothing. Since then both "env": {"CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION": "true"} and "promptSuggestionEnabled": true were added to ~/.claude/settings.json (JSON validated), followed by a full app restart and a full machine reboot. Still zero suggestions.
• No error output anywhere; the feature is just silently absent.

Environment

• Version: 2.1.207
• Platform: Windows 11 Home (10.0.26200)
• Interface: Claude Code desktop app
• Model in use when it stopped and since: Opus 4.8 and Sonnet (occurred on both)

Repro conditions on this machine (100% reproduction of the absence)

1. Any session, any project, any model, 2026-07-12 onward
2. Do substantive multi-turn work; end turns with clear next-step recommendations
3. No suggestion ever renders

Happy to provide logs or run a debug build if pointed at what to capture.