Omniscio

OmniscioGuidesClaude Code prompt suggestions

Claude Code prompt suggestions: on, off, and missing

Claude Code can put a faded suggestion in the prompt box before you have typed anything: an example when a session opens, and a guess at your next prompt after Claude answers. When it works, Tab takes it. When it stops, the cause is one of a handful of documented conditions, and only some of them live in the setting with the matching name. This page covers what the suggestion is, what suppresses it, which switch wins, and how to turn it off for a team.

Sources read October 27, 2026. By the Omniscio team. Updated .

Find the situation you are in

The next-prompt suggestion is a small, documented feature with a short list of reasons to be missing, so most readers find the answer on this page. If the request never reaches a model at all, or the session will not start, this is the wrong page; the connection errors guide owns that family. If the Code tab around the prompt box is the confusing part rather than the prompt box itself, the sidebar guide owns that surface.

What a prompt suggestion is, and what writes it

Claude Code shows two versions of the same idea. When you first open a session it puts a grayed-out example command in the prompt input, picked from your project's git history, so the example tends to involve files you have been working on. After Claude responds, it can suggest your next prompt based on the conversation, such as a follow-up step from a multi-part request or a natural continuation of what you were doing.

For the next-prompt suggestion, press Tab or Right arrow to place it in the prompt input, then Enter to submit it. Start typing and it goes away. The keyboard table describes Tab as accepting an autocomplete suggestion while suggestions are showing, which is the same key, doing a smaller job.

Each next-prompt suggestion is generated by a short background request to the same model your session is using. That request counts toward your plan's usage limits or your API costs, and because it reuses the conversation's prompt cache it is mostly cache reads plus a few output tokens, so the added cost is small.

One thing sits near it and is not it. This is not inline code completion: the desktop app does not offer autocomplete-style code completions at all, and works through conversational prompts and explicit code changes instead.

Why they stopped appearing

Work down this list in order. The first two rows are the ones an install, an upgrade or a provider change can produce, and neither of them is fixed by the setting whose name matches the symptom.

What you seeThe documented causeWhat to do
Nothing at all, in the first session after an install or upgradeClaude Code had not fetched its feature flags yet, so a flag-gated feature can be missingStart another session on the same machine; the flags were saved during the first one
Nothing at all, on Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or a Claude apps gatewayThose sessions skip the feature-flag fetch, so suggestions are left off by default and the Prompt suggestions toggle is hidden in /configSet CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=1 if you want them there
Nothing at all, with telemetry turned offDISABLE_TELEMETRY and the other traffic variables also stop the feature-flag fetchSame: the setting has no effect, and only the environment variable turns them on
A suggestion on some turns and not othersDocumented skips: a cold prompt cache, the first turn in some sessions, a previous response that ended in an error, and plan modeNothing to fix; these are deliberate
Suggestions, but fewer of them, with a noticeYou left many suggestions unused in a row, so Claude Code throttles themUse one, or set the environment variable below
None in your teammate's session in an agent teamTeammates' sessions skip suggestions by default; the lead's session shows themNothing to fix; this is the documented default

The usage-limit skip has its own escape. Claude Code skips a suggestion while your account is close to or at its usage limit, and setting CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION to true keeps that skip from applying until you actually reach the limit. The variable itself needs v2.1.238 or later, and before that release Claude Code skipped suggestions near the limit even with it set, so an older build will look like it ignored you.

The notice is worth recognising, because it is not an error and nothing is broken. Showing fewer prompt suggestions · use one to bring them back means Claude Code is showing suggestions less often because you left many in a row unused. Using one returns you to the usual frequency, and so does setting the environment variable to true.

Two setups make the first row a permanent state rather than a one-off. If every run starts in a clean environment, such as a CI container without the flags an earlier session saved, then every run is a first session. And if you authenticate with ANTHROPIC_AUTH_TOKEN and no API key against a host other than Anthropic's, such as an LLM gateway, Claude Code has no credential to fetch the flags with.

Then there is the plain case: somebody turned it off. The setting is promptSuggestionEnabled, it defaults to true, and the Prompt suggestions toggle in /config is what writes it. If a colleague set it to false, or a managed settings file did, no amount of restarting brings the gray text back until that value changes.

The three ways to change it, and which one wins

There are three places to change this, and they are not equal. The /config toggle writes promptSuggestionEnabled in your settings file, and that key defaults to true, so an empty settings file means suggestions are on. The environment variable CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION is per session and takes precedence over the setting for that session, so an exported true overrides a settings file that says false.

To turn suggestions off entirely there are three routes, and any one of them is enough: turn off Prompt suggestions in /config, set promptSuggestionEnabled to false in your settings file, or set the environment variable to false.

For an organisation, one setting is not enough. Set promptSuggestionEnabled to false in managed settings, and also set CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION to false under the managed env key, otherwise a user can re-enable suggestions with their own environment variable.

One caveat matters more than the rest. Prompt suggestions need a claude.ai or Console account with telemetry on. On Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, or with telemetry turned off such as by DISABLE_TELEMETRY, the promptSuggestionEnabled key has no effect, and only CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=1 turns suggestions on.

Where you can reach that switch depends on the surface. /config is part of the terminal interface: the VS Code chat panel and the desktop app do not open it, so in those you change a settings file or use the app's own settings screen.

Print mode does not generate suggestions by default. To get them, pass --prompt-suggestions together with -p, --output-format stream-json and --verbose, and Claude Code emits a prompt_suggestion message after each turn that generates one. The same skips apply on this path: short conversations and cold prompt caches can produce no message at all, so a single short query is not a test of whether the flag works.

claude -p --prompt-suggestions --output-format stream-json --verbose "summarise this repo"

For a team running automations, the route that holds for every user on the machine is a managed settings file, with the matching value under the managed env key. If a script is producing no prompt_suggestion messages, check the flags first: without all three of -p, the stream-json output format and verbose mode, the flag is not in play.

The desktop Code tab

In the desktop app's Code tab, after Claude replies, a suggested next prompt can appear as gray text in the empty prompt box. Press Tab or Right arrow to place it, edit it if you want, then Enter to send. Pressing Enter before you accept the suggestion does not send it, and the suggestion shows only while the prompt box is empty and has no attached files, so typing or attaching a file makes it disappear.

The switch lives in that app's own settings rather than in /config, which the desktop app does not open: go to Settings > Claude Code and turn off Prompt suggestions under Sessions. That stops suggestions in each session from the next time it starts or resumes. The VS Code chat panel has no /config either, so there the setting is one line in your settings file.

What other people have reported

Four reports are open in Anthropic's issue tracker as we write this, and none of them is an Anthropic statement about the feature. They are worth reading because they describe a symptom the documented list above does not explain: suggestions that simply stop after an update, with the setting on and nothing in the environment.

  • Issue 103335, open since 2026-10-19: in the desktop app's Code tab, prompt suggestions stopped appearing around the 2026-10-13 update. 16 reactions, 15 comments.
  • Issue 102208, open since 2026-10-14: suggestions never appear on Windows in both the terminal and the desktop app, with the setting showing on. 7 reactions, 6 comments.
  • Issue 102227, open since 2026-10-15: a collection of desktop redesign complaints that includes the missing suggestions. 7 reactions, 4 comments.
  • Issue 97410, open since 2026-09-26: a feature request, not a bug, asking for prompt suggestions in the mobile app's Remote Control view. 6 reactions, 5 comments.

Nothing on this page claims those reports are fixed. If you are in that situation, work the documented list first, because a setting or a feature flag is something you can change in a minute, and then read the reports to see whether your version matches theirs.

Where this page stops

This page is about the suggestion text in the prompt box and the switches that control it. Three related subjects have their own pages, and this one does not restate them.

  • The Code tab's sidebar, its grouping, its sorting and its archive behaviour belong to the sidebar guide, which is live.
  • The cache a suggestion reuses is the subject of the prompt caching guide, which covers what is cached, for how long, and what the read tokens cost.
  • Your own reusable / commands are a different feature, covered by the custom slash commands guide.

The You should know note looks similar and is not this: it is a built-in mod that requires an explicit opt-in, and it comments on Claude's work rather than guessing your next prompt.

Omniscio and prompt suggestions

Omniscio is the desktop app we make for running many Claude Code sessions at once. It does not predict your next prompt, and nothing in the app writes or replaces the suggestion described on this page: every session it starts runs Claude Code in print mode with stream-json output, and the app does not pass --prompt-suggestions, so the suggestion is not generated inside an app session at all.

What the app does have is a different helper with a similar-sounding name. The composer can show three short replies above the input when an agent asks you a question; they are switched off by default, generated on a small utility model rather than your main session, and their job is to answer an agent's question, not to guess your next prompt.

Separately, the app has an in-development Writer Studio, behind a Lab toggle, with its own inline autocomplete that shows ghost text as you type and is also off by default. It belongs to the Writer and is unrelated to Claude Code's prompt suggestions. We are naming both so that "Omniscio has no prompt suggestions" is not read as a claim about ghost text anywhere in the app.

If you are running several sessions at once, the features page lists what the app does and the download page has the installers.

Sources

Every documented behaviour on this page rests on one of these pages, and each was read on the date beside it.

  • Interactive mode, for the prompt suggestion section, the skips, the notice, the print-mode flag and the way to turn suggestions off. Read October 27, 2026.
  • Settings reference, for promptSuggestionEnabled, its default, its scope and the account and telemetry requirement. Read October 27, 2026.
  • Environment variables, for CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION and for the feature-flag fetch it depends on. Read October 27, 2026.
  • CLI reference, for --prompt-suggestions and the three flags it requires. Read October 27, 2026.
  • Claude Code in the desktop app, for the Code tab's suggested prompt, where its setting lives, and the note about inline code completions. Read October 27, 2026.
  • Settings, for /config being terminal-only and for where the other surfaces change settings. Read October 27, 2026.
  • Issue 103335, open, for the report that suggestions stopped in the desktop Code tab after the 2026-10-13 update. Read October 27, 2026.
  • Issue 102208, open, for the Windows report with the setting on. Read October 27, 2026.
  • Issue 102227, open, for the desktop redesign collection that includes the missing suggestions. Read October 27, 2026.
  • Issue 97410, open, for the mobile Remote Control request. Read October 27, 2026.
  • One capture run on this machine on October 27, 2026: claude --version, which printed 2.1.311 (Claude Code).

This is a young part of the product, and the pages above change with releases. If a release note mentions suggestions, re-read them before trusting a copy of this page from last month.

Common questions

Why did my Claude Code prompt suggestions disappear?

Start with the session, not the setting. Claude Code leaves prompt suggestions off by default in a session that has not fetched its feature flags, which means a session on a third-party provider or through a Claude apps gateway, a first session after an install or upgrade, and any session where telemetry is turned off. If none of that applies, the other documented causes are the switch itself, a usage limit, a cold prompt cache, plan mode, or a previous turn that ended in an error.

How do I turn Claude Code prompt suggestions off?

Turn off Prompt suggestions in /config, set promptSuggestionEnabled to false in your settings file, or set CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION to false, which takes precedence over the setting. Anything typed or exported is for that shell or session only. To turn them off across an organisation, set promptSuggestionEnabled to false in managed settings and set the same environment variable to false under the managed env key, so nobody can switch them back on with their own variable.

Do Claude Code prompt suggestions cost extra?

They are not free, but they are small. Each suggestion is generated by a short background request to the same model the session is using, and the request counts toward your plan's usage limits or your API costs. Because it reuses the conversation's prompt cache, it is mostly cache reads plus a few output tokens. Turning suggestions off removes that request entirely.

Why is there no prompt suggestion after my first message?

Several documented skips cover that case. Claude Code skips the suggestion when the prompt cache is cold, after the first turn of a conversation in some sessions, when the previous response ended in an error, and while you are in plan mode. In an agent team, teammates' sessions skip suggestions by default and the lead's session shows them.