Skill or setting to skip permission prompts for esoteric/low-risk commands

Opening this to brainstorm.

Problem

Claude Code in the terminal prompts for permission on commands that feel esoteric or low-risk in context — compound commands (a && b), piped bash, flag variations, and shell idioms that the user has effectively already approved in spirit. Each new variant re-prompts, which breaks flow.

Ask

A skill or setting that lets the user mark certain classes of commands as "don't ask" — for example:

• a broader allowlist pattern language (globs/regex over full command strings, not just the binary)
• a "trust compound forms of already-allowed tools" mode
• a skill the user can invoke to teach Claude which shapes are safe for this project

Open questions for discussion

• How to keep this safe (scope to project? session? expire?)
• Should it live in settings.json permissions, or as a dedicated skill?
• How to handle compound commands where one half is allowed and the other isn't?
• Can Claude learn from prior approvals in a session without a config change?

Not a bug — a UX discussion.