Insight and code workflows

Insight and code workflows are the commands and views that show what a session did and what it used: its changes, its context, usage and status, the files you name with @, its exported transcript, and going back with /rewind. They work with every agent unless a section says otherwise.

/diff and the Git view

/diff opens the Git view on this session’s repository: its branches, its history, and its working changes with the commit box. See Git and Files. It needs a git repository.

Each turn’s footer in the transcript says how many files the agent edited, with the lines added and removed.

Naming files with @

Type @ at the start of a word to pick a file or a folder from the session’s folder. Keep typing to narrow the list: the letters you type match in order, not only side by side, so rdme finds README.md. The best matches come first, at most 50 of them, and folders end with /. The list says “Finding files…” while it looks, and “No matching files” when nothing matches. Arrow keys move through it, Enter or Tab picks, and Escape closes it.

Picking inserts @path into your message, in quotes when the path has a space. Claude Code reads the file itself; Codex and Gemini CLI get the path and read it with their own tools.

A path that contains a double quote can’t be picked.

/export

/export saves the transcript as a Markdown file. You choose where to save it. The Export the transcript button in the transcript panel’s header does the same.

• At the top: the session’s title, its agent, its folder, a worktree session’s branch, and when it was exported.
• Then the messages, the thinking, each tool with its output and changes, each subagent’s work (collapsed), the permissions you answered, the questions with your answers, the plans with the feedback you sent, the forms and links, and the status rows.
• The files you sent with your messages aren’t in it, only the words. Events Moontors didn’t recognize are a count.

Context, usage and limits

• Context: a small ring shows how full the agent’s context is, on the card and beside Send in the composer. It fills as the context does, and changes colour past 80%. Hover over it for the share used and the tokens used of the context’s size. Where the agent can compact, a click on the composer’s ring compacts the conversation now, as /compact does. /context takes you to the composer’s ring. Claude Code and Codex report it; Gemini CLI doesn’t, so it shows no ring.
• Usage: the transcript panel’s header shows the tokens in and out, and the cost where it’s a charge. A forked session’s cost includes the history it was forked from. With a Claude subscription, the cost isn’t shown, since the agent’s figure is an estimate at API prices, not what you pay.
• Usage limits: your account’s usage limits and when they reset, with a Claude subscription or a ChatGPT sign-in for Codex.

/status

/status opens the Configuration dialog’s Status tab: the app’s version and when it was built, the agent, the model, mode and folder, a worktree session’s branch, the conversation’s id, how full the context is, the session’s usage, your account’s kind and plan, and your usage limits. At the bottom are the Third-party notices, Terms of Use and Privacy Policy buttons.

While the session’s agent is running, the tab also shows what the agent says of itself:

• Claude Code: its version, and the instruction files and MCP servers in effect.
• Codex: its version, and the instruction files and MCP servers in effect. Its Managed row says “Your organization’s requirements apply” when they do. Its Config layers list Codex’s config layer stack, lowest precedence first, each layer named as Codex’s own /debug-config names it, and shown enabled, or disabled with the reason.
• Gemini CLI: its version, and its About: the CLI version, the model, how you signed in (Auth Method), and the Google Cloud project when one is set.

With Codex, Warnings lists the warnings Codex sent during the session, latest last.

The agents’ own commands for these open the same tab: Claude Code’s /usage, /cost and /stats; Codex’s /status, /usage, /pwd or /cwd, /warnings and /debug-config; and Gemini CLI’s /about, /stats and /usage.

/rewind

/rewind goes back to before one of your prompts. Pick the prompt from the list, or use the rewind action on the prompt itself. Moontors says how many turns the agent will forget, and asks you to confirm.

• The list says “Reading the whole transcript…” while it loads, then asks “Rewind to before which prompt?”. A prompt you can’t rewind to is dimmed, with the reason under it.
• The confirmation names the prompt and how many turns the agent forgets, with Cancel and Rewind.
• With Claude Code, a checkbox offers to restore the files the agent changed since then, with how many and the lines added and removed. It says “Finding which files it would restore…” while it looks, “It changed no files to restore.” when there are none, and “Couldn’t tell which files it would restore.” when it can’t tell.

	Conversation	Files
Claude Code	yes	yes, if you choose
Codex	yes	no
Gemini CLI	no	no

• With Claude Code, you can also restore the files the agent changed since that prompt. Only the agent’s own edits are restored, not changes made through shell commands. Restoring also overwrites any changes you made to those files by hand since then.
• With Codex, only the conversation goes back; the files stay as they are.
• The turns after the prompt stay in the transcript, dimmed as undone, and the prompt comes back to the composer, unless you’ve started writing there.
• Not while a turn is running, and not while the agent’s background tasks are running, as it would stop them: stop them first.
• Not to a prompt before the last /clear, nor to one an earlier rewind already went back past.
• With Claude Code, you can’t rewind past a point where the conversation was compacted.
• Claude Code keeps its file backups for 30 days by default (its own cleanupPeriodDays setting).