Feature request: extensible prompt suggestions (hooks/config/CLAUDE.md)

Summary

Claude Code's prompt suggestions (the grayed-out next-message autocomplete in the input area) are great, but currently appear to be closed — no settings, hooks, or CLAUDE.md directives to steer them.

Use case

I maintain a custom catalog of slash commands and skills (research workflows, lit-review, identification strategy, manuscript audit, etc.). I've been surfacing context-aware skill suggestions in my powerline statusline as a workaround. Native prompt-area suggestions would be a much better surface for this — but only if I can feed in my own logic about which commands/skills are relevant given the current conversation.

Proposed extension points (any one would unblock this)

1. Hook event — e.g. UserPromptSuggestion that fires when suggestions are generated, with the conversation context, allowing the hook to return additional candidate suggestions.
2. Settings.json key — e.g. promptSuggestions: { sources: [\"slash-commands\", \"recent-skills\", { command: \"./bin/my-suggester.sh\" }] } that lets users plug in custom suggestion generators.
3. CLAUDE.md directive — e.g. a ## Prompt suggestions section listing slash commands or skills that should be considered when contextually relevant.

Why hooks are insufficient today

Existing hooks fire on tool execution and prompt submission, not on suggestion generation. There's no event surface for the suggestion engine.

Workaround currently used

Statusline (powerline) renders contextual skill hints — works, but separate from the input field where suggestions actually need to land.