Permission Prompt and doc Gap

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Feature request and doc gap: The permission prompt "Yes, don't ask again" button writes a pattern to settings.local.json using an algorithm that strips to the leading binary. For example, running "git status -s" generates a Bash pattern matching all git commands, not just git status. This is:

(a) undocumented - the settings.json schema describes the pattern format but not how the UI generates patterns from approvals,
(b) one wildcard-step broader than users typically intend, creating silent over-permissioning, and
(c) not editable - the UI appears to offer no way to narrow the suggested pattern before accepting.

Requests:

1. Document the pattern-generation algorithm in the Claude Code settings docs.
2. Add an edit-pattern or narrow-pattern option to the permission prompt UI.
3. Consider defaulting to subcommand-preserving patterns rather than binary-only patterns. Subcommands carry semantic intent that arguments do not.

Proposed Solution

Three options, any of which would address the issue:

1. Document the pattern-generation algorithm in the Claude Code settings docs, so users know what pattern will be written when they click "Yes, don't ask again."

1. Add an "edit pattern" field to the permission prompt UI, letting users narrow the suggested pattern before accepting. Default the field to the auto-generated pattern, let users tighten it.

1. Change the pattern-generation default to preserve the subcommand, not just the binary. For "git status -s" generate a pattern matching "git status <args>" rather than "git <anything>". Subcommands carry semantic intent that arguments do not.

Option 2 is the most flexible. Option 3 is the lowest-effort fix with the biggest safety win - no UI change, just a different default.

Alternative Solutions

_No response_

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_