### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? Contextual prompt suggestions (grayed-out follow-up suggestions appearing in the input box after each Claude response, based on the current conversati

[BUG] Contextual prompt suggestions during conversations stopped working (regression ~March 1-2)

Status Open

Reported on v2.1.68

Maintainer reply None cached

Activity 7 comments · opened Mar 4, 2026

Preflight Checklist

[x] I have searched existing issues and this hasn't been reported yet
[x] This is a single bug report (please file separate reports for different bugs)
[x] I am using the latest version of Claude Code

What's Wrong?

Contextual prompt suggestions (grayed-out follow-up suggestions
appearing in the input
box after each Claude response, based on the current conversation
context) stopped
working approximately March 1-2, 2026.

This feature was working reliably for a long time before that. It is
distinct from the
basic "new session" prompt suggestion — it generated contextual
follow-ups throughout
the entire conversation.

## Environment

Claude Code version: 2.1.68
OS: Ubuntu 24.04 (Linux)
Connection: SSH session
Subscription: Claude Max

What Should Happen?

After each Claude response, grayed-out suggestions should appear in
the input box
based on the current conversation context (not just git history
suggestions at session
start).

Error Messages/Logs

Steps to Reproduce

Open terminal and start Claude Code: claude
Type any message and send it (e.g. "help me refactor this

function")

Wait for Claude to finish responding
Observe the input box — no grayed-out contextual suggestion appears
Repeat for several back-and-forth turns — suggestions never appear

## What used to happen (before ~March 1-2)

After each Claude response, a grayed-out suggestion would appear in
the input box
reflecting the current conversation topic (e.g. if discussing
refactoring, it might
suggest "add tests for this function" or "explain what this does").
Pressing Tab would accept the suggestion.
This worked consistently throughout entire conversations, not just at
session start.

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.68

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

_No response_

View original on GitHub ↗

7 Comments

rolandtolnay · 5 months ago

I really enjoyed this feature and used it often, any ETA on a fix?

mderzi · 5 months ago

Claude Code v2.1.71 and still having the issue, any workarounds?

doronzeltzer-hub · 5 months ago

Prompt suggestions not working
▎ - Version: 2.1.78, Windows 11, Git Bash
▎ - tengu_crystal_beam.budgetTokens is 0 (fetched fresh from server)
▎ - Feature worked previously, stopped at some point
▎ - promptSuggestionEnabled: true in settings, tengu_prompt_suggestion: true in feature flags
▎ - totalSpeculationTimeSavedMs: 0 — suggestions have never fired

Investigated locally and found that tengu_crystal_beam.budgetTokens is set to 0 (confirmed fresh from server, not stale cache). This controls the speculation token budget that powers suggestions. With budget=0, suggestions
never fire — totalSpeculationTimeSavedMs in stats-cache.json is 0 for all time. Version 2.1.81, Windows 11, Claude Pro subscription. Feature was working previously.

shanemhamilton · 4 months ago

Still broken in 2.1.101 (latest). The showSuggestedNextSteps setting no longer exists in the compiled CLI — adding it to ~/.claude/settings.json is a no-op.

This was one of the most useful ambient features of the TUI during long agentic sessions: the dimmed hint in the input footer surfaced the obvious next action (run tests, review the diff, commit) without forcing a context-switch back into my own head. Losing it has measurably slowed down multi-step workflows where I step away and come back mid-task.

+1 on restoration — even a lightweight rotating set of static suggestions gated by the original flag would be a big win over the current static ? for shortcuts.

realdimas · 4 months ago

Forcing via the env var works for me (although the suggestions aren't as good as they used to be):
CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=1.

DieterReichelt · 2 months ago

+1 — this regression removed something I relied on. After Claude finished a task, the contextual ghost-text hint pre-empted the obvious next action (commit and push, deploy to the next server, etc.), which kept momentum on multi-step and homelab/multi-server workflows. Without it I have to re-derive the next step manually every time. Please restore it — ideally with a toggle.

inth3shadows · 2 months ago

For anyone who restored suggestions with CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=1
but still sees them come and go: there appears to be a separate, still-live cause
that looks distinct from the March regression discussed here (which, per the
reports above, seems server-side — budgetTokens=0 / a flag change).

In my testing, inline suggestions are suppressed whenever the
anthropic-ratelimit-unified-status response header reads allowed_warning (i.e.
when you're near your weekly usage limit), and return once it resets to allowed
— the session stays fully usable throughout, so it's easy to misattribute. I've
filed this separately as #72495. I can't say for certain it's the same thing
everyone here is hitting, but it may explain the intermittent "works some days,
not others" reports.