Shell command history

한국어 · Available since v0.6.0

Ordinary local Bash and Zsh sessions can put human commands back in the shared timeline. Commands are observed at the shell’s execution boundary. Conn does not reconstruct them from keyboard input, Enter presses, terminal text or prompt patterns.

Use

Select a local Bash or Zsh profile in Settings → Profiles and open a new tab. Run a command, then open Timeline to see its entry. Existing sessions need a new tab; user startup files are not edited.

One entry moves from Running to Completed, Failed, or Completion unconfirmed. Expand it for the working directory, exit code and elapsed time. The exit code describes the outer shell command/list, not every pipeline process or a background job’s eventual result. An agent command is joined to its existing request/approval entry by ID. Uncertain attribution is labelled Shell.

Action	Timeline	Agent screen/control
Run a command at the integrated local prompt	One lifecycle entry	Existing mode and permissions
Type a password into sudo/SSH or a program	No input entry	Existing screen permissions; visible output is still visible
Open Vim, a REPL, nested shell, or tmux	Outer command only	Existing mode and permissions
Connect with ssh host	The local SSH command, ending when it returns	Continues inside SSH; no remote helper is installed
Use an external private session	No history and no hooks	Unavailable while private. The owner sharing action (v0.7.0 and later) enables collaboration in the same session; human takeover alone does not

Startup compatibility

• Bash: an initial PROMPT_COMMAND bootstraps hooks after normal startup. Existing prompt commands and their exit status are preserved. An existing DEBUG trap, function tracing or extended debugging prevents installation. Startup files that overwrite PROMPT_COMMAND can prevent the bootstrap and leave command recording unavailable. Bash uses fresh history entries, so disabled history, history exclusions and suppressed duplicates can leave commands unrecorded.
• Zsh: a temporary .zshenv restores the original ZDOTDIR, sources the original .zshenv, and arranges first-prompt installation of preexec/precmd hooks. Other startup files and existing hook arrays remain in use. Startup scripts that replace these hooks can prevent recording.
• Initial support is for interactive local profiles with default/login/interactive flags. Custom startup commands, remote backend profiles, other shells and Windows report unavailable. There is no fallback to keystroke capture.

Only the initial shell is integrated. Installing remote hooks, broadening shell support, and detecting/redacting secrets embedded in command arguments are separate work. An integration does not certify command arguments as safe to share.

Delivery and verification

Lifecycle events use a private, bounded temporary mailbox; no helper executable or network service is needed. There are at most 64 queued records, each bounded to 64 KiB. Overflow or invalid framing stops recording for that tab; open a new tab to reconnect. Normal shutdown removes the mailbox. A process crash may leave temporary command data, subject to the same-user boundary in security.md.

crates/core/tests/shell_integration.rs exercises real Bash/Zsh PTYs, authentication input, nested shells, private sessions, existing prompt/debug hooks, suppressed history, agent correlation and unconfirmed completion. Frontend tests cover live and saved lifecycle records. Zsh can be selected with CONN_TEST_ZSH. The separate authentication fixture project covers real SSH password/retry/key authentication, hidden/masked/visible prompts and Vim. These Linux checks do not establish native macOS shell-hook behavior; platform-specific hands-on verification remains necessary. See the v0.6.0 validation limits.