[FEATURE] Let users steer prompt suggestion content (CLAUDE.md, output styles, or a hook)
Problem
Prompt suggestions are currently all or nothing. The only control is enablePromptSuggestion. The generation itself is a harness-side background request that reuses the conversation's prompt cache, and there is no way for a user to influence what gets suggested.
In practice the suggestions are hit or miss. They also appear inconsistently (skipped on cold cache, plan mode, first turn), which is documented, but when they do appear there is no way to make them more relevant to how I actually work.
Proposal
Give users some way to steer suggestion content. Any of these would help, roughly in order of preference:
- Respect CLAUDE.md or a dedicated settings key with short guidance for the suggestion generator, e.g. "prefer verification and shipping steps over exploratory follow-ups", "suggest slash commands from my installed skills when they fit".
- Let output styles apply to suggestion generation, so a style that shapes responses can also shape suggested follow-ups.
- A hook point that receives the generated suggestions and can filter or rewrite them before display.
Why this matters
The suggestion request already sees full conversation context via the cached prompt, so the machinery for context-aware suggestions exists. What is missing is a small amount of user-provided steering. Teams with established workflows (custom skills, verification gates, PR conventions) would get far more value from suggestions that point at the next step in their actual workflow instead of generic follow-ups.
Related issues
Not a duplicate of the existing prompt suggestion requests: #71788 and #82065 ask for disabling, #74826 asks for frequency control, #77870 asks for multiple candidates. This one is about controlling the content.