Feature: `!!` bash-mode variant — run command AND have Claude respond to its output

Problem

! <command> runs a shell command in-session and injects its output into the conversation. Whether Claude responds to that output is controlled by the global respondToBashCommands setting (default true since v2.1.186; I run with false). But the intent genuinely varies per command within one session:

• Auth logins, quick fixes: context-only is right.
• "Run this and deal with the result" (a failing build, a curl probe, a script the agent staged for me because a permission rule blocked it from running the command itself): I want Claude to act on the output immediately.

With respondToBashCommands: false I have to type a second "respond to that" message for the second case; with true, every incidental command produces response noise. There is no per-invocation override in either direction.

Proposal

• ! <command> — run, inject output as context, no response (quiet default).
• !! <command> — run, inject output, and have Claude respond/act on it (the marked, exceptional form).

Ideally the single-bang behavior keeps honoring respondToBashCommands for backwards compatibility, and !! forces a response regardless.

Real-world motivation

Long infra session where the permission classifier (correctly) blocked the agent from credential-setting and data-injection commands; the agent staged scripts and I executed them via !. Every single time, the natural next step was for the agent to verify/continue based on the output — which today needs an extra "ok now respond to that" message. !! would collapse that round-trip.

🤖 Filed via Claude Code on the user's behalf.