Autocomplete suggestions should not propose hidden dot-files or visibility changes

Summary

Gray autocomplete/suggestion text in Claude Code can be accepted with tab+enter. Currently, these suggestions can include actions that are destructive, irreversible, or violate user principles — and are too easy to trigger accidentally.

Examples observed

1. Hidden dot-files: Suggestion to create .submodules (dot-prefix file)
2. Repository visibility: Suggestion related to changing GitHub privacy settings

The concern

Users working quickly may tab+enter without fully reading the gray suggestion text. When that suggestion creates a hidden file or changes security/visibility settings, the accidental trigger can:

• Create files the user didn't intend (especially hidden ones that are easy to miss)
• Change repository visibility in ways that are hard to notice and reverse
• Violate project conventions (some teams have "no hidden files" policies)

Proposed solutions

1. Exclude from autocomplete: Don't suggest creating dot-prefix files or changing visibility settings
2. Require confirmation: If these suggestions remain, require explicit confirmation before executing
3. Visual differentiation: Make potentially destructive suggestions visually distinct from safe ones

Context

This came up during a working session where the user has explicit project canon: "nothing hidden" (no dot-files) and "private stays private." The autocomplete suggestions directly contradicted these principles and could have been triggered accidentally.

Thank you for considering this feedback.