[FEATURE] Auto-write session log when running with --dangerously-skip-permissions

Resolved 💬 3 comments Opened Mar 12, 2026 by SwizzzleStyx Closed Apr 12, 2026

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

When you run Claude Code with --dangerously-skip-permissions, there is no record of what happened. If Claude creates, modifies, or deletes something unexpected, you have no way to review it after the fact unless you happen to be in a git repo and caught it in time.

The longer and more autonomous the session, the worse this gets.

Proposed Solution

When --dangerously-skip-permissions is active, automatically write a session log to .claude/bypass-session-YYYY-MM-DD-HHMMSS.log that records:

  • Files created, modified, or deleted (with full paths)
  • Shell commands executed
  • Timestamps for each action

No configuration needed. Just always on when bypass mode is active. The log files can be gitignored by default.

Note: this is related to but distinct from #33141, which proposes per-command audit logging as a permission level. This is specifically a session-level log that fires automatically in bypass mode.

Alternative Solutions

Currently the only option is to manually track changes through git, which only works if the directory is a repo and you remember to check before and after. There is no built-in way to see what Claude did during a bypass session.

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

I launch Claude Code with --dangerously-skip-permissions to work through a long coding task without interruptions. The session runs for 20 minutes and touches a lot of files. Afterward I want to review exactly what changed, but short of running git diff I have no idea what happened. A log file at .claude/bypass-session-2026-03-12-091700.log would let me open it and see a clean list of every file touched and every command run.

Additional Context

_No response_

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗