Add a frequency / always-on mode for prompt suggestions

Open 💬 0 comments Opened Jul 6, 2026 by Evan-Haug

What

Prompt suggestions (the ghost-text next-prompt guess in the input box) are currently a binary on/off via CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION / the /config toggle. This requests a way to make them appear more often, up to an "always attempt" mode.

Why

When they show up they're genuinely useful, but in practice they appear rarely. Per the Interactive mode docs, generation is skipped when:

  • the prompt cache is cold,
  • after the first turn of a conversation,
  • in plan mode.

The cold-cache skip in particular means suggestions silently disappear across much of a normal working session, and there is no user-facing lever to trade a little cost for more suggestions. Some of us would happily pay the extra tokens to see them consistently.

Proposed

A setting (env var and/or /config) with tiered behavior, e.g.:

  • off — current opt-out
  • default — current cache-warm-only behavior
  • always — attempt a suggestion every turn even on a cold cache, accepting the extra token cost

Note on the docs

The Interactive mode docs currently contain an apparent contradiction — "suggestions continue to appear" after Claude responds vs. "automatically skipped after the first turn of a conversation." Clarifying the real trigger conditions would help regardless of whether the frequency knob lands.

Alternatives considered

  • --prompt-suggestions already exists but is print-mode / stream-json --verbose only — it emits a prompt_suggestion message for scripting, not the interactive typing experience.

Related

Distinct from #17572 and #16032, which both request disabling suggestions (both closed as duplicate). This is the inverse: making them more frequent.

View original on GitHub ↗