How It Works
When prompt suggestions are enabled, the Claude Code agent generates a predicted next user prompt at the end of each turn. The suggestion appears as a clickable chip below the agent’s response in the workspace input area. Suggestions are context-aware — they take into account the full conversation history, the agent’s last action, and the current state of the workspace. Common suggestions include follow-up tasks, test requests, or next steps in a multi-step workflow.Prompt suggestions piggyback on the parent turn’s prompt cache, making them nearly free in terms of token cost.
When Suggestions Appear
Suggestions follow these delivery rules:| Condition | Suggestion shown? |
|---|---|
| After a normal agent turn | Yes |
| First turn of a session | No |
| After an API error | No |
| In plan mode | No |
| Feature disabled | No |
Enabling Prompt Suggestions
Prompt suggestions are configured in the Claude Code agent settings. When enabled, suggestions are generated automatically for all workspace sessions using Claude Code. The feature can also be controlled via environment variable:false to disable suggestions entirely without changing the in-app setting.
Usage Tips
- Quick iteration — after the agent makes code changes, the suggestion often proposes running tests or reviewing a related file
- Workflow continuity — suggestions help maintain momentum by proposing logical next steps
- Ignore freely — suggestions are non-blocking; type your own prompt anytime to override
Prompt suggestions are only available for the Claude Code agent runtime. GitHub Copilot workspaces do not support this feature.