The shim always injected --session-id with a fresh UUID, which broke
`claude --resume <id>` and `claude --continue` by conflicting with the
user's session flag. Now scans args and skips injection when the user
already specifies a session/resume flag.
Also passes through subcommands (mcp, config, api-key) without injecting
hooks or session flags since they don't support them.
Instead of creating a merged config directory and injecting
CLAUDE_CONFIG_DIR on every terminal spawn, place a thin wrapper
script at Resources/bin/claude that intercepts claude invocations
to inject --session-id and --settings flags. This eliminates
blocking I/O on terminal creation and removes config management
complexity.
- Add Resources/bin/claude wrapper script with hook injection
- Add shell integration PATH fix (re-prepend after .zshrc/.bashrc)
- Add transcript reading for richer stop notifications
- Add set_status/clear_status to notifications socket allowlist
- Add Settings toggle to disable Claude Code integration
- Update docs to reflect automatic integration approach
- Unset CLAUDECODE env var to avoid nested session detection
