Slash Commands Reference

AI Workflow · mixed

Slash commands are typed in the input bar and are dispatched before normal natural-language classification. The canonical list below is generated from the app's own command dispatch, so it always matches the current build.

These commands are generated from the dispatch block in InputModel.swift.

/fix [follow-up] (Free): Reads the last command failure and proposes an editable correction. A follow-up after /fix answers an AI clarification.
/explain [command or text] (Free): Explains the last command/output, or the supplied command/text. It is read-only and does not run anything.
/agent <task> (Pro): Starts a manual multi-step agent run. AITerm asks for each step and runs it only after approval.
/auto <task> (Pro): Starts Autopilot. Safe in-project steps can run automatically; risky steps pause for approval. Requires an integrated zsh/bash shell.
/runbook save-last <N> (Pro): Captures the last N executed commands into a runbook draft. N must be a positive integer.

/fix (Free)

/fix looks at your last command that failed, reads its exit code and the surrounding context, and proposes an edited command that should work. The fix arrives as a normal editable proposal — you still approve it (and it still passes the security gate) before it runs. If the AI needs a clarification, type a short follow-up after /fix to answer it. There's also a keyboard shortcut, ⌘⇧F.

/explain (Free)

/explain is read-only. With no argument it explains your last command or its output; give it a command or some text to explain that instead. It streams an explanation into a panel and never runs anything. Shortcut: ⌘⇧E.

/agent (Pro)

/agent <task> starts a manual, step-through agent run: AITerm proposes one command at a time and runs each only after you approve it. It works on any shell. See The Autonomous Agent.

/auto (Pro)

/auto <task> starts Autopilot: provably-safe in-project steps can run automatically, while risky steps pause for your approval. Autopilot needs an integrated zsh/bash shell so AITerm can track when each command finishes. Your Safety Profile and the security gate still apply to every step.

/runbook save-last N (Pro)

/runbook save-last <N> captures the last N executed commands into a runbook draft and opens the editor. N must be a positive integer, for example /runbook save-last 5. See Runbooks.

Empty or invalid forms

/agent with no task shows usage for step-by-step mode.
/auto with no task shows usage for autonomous mode.
/runbook without the save-last <N> form shows usage with an example.

/fix and /explain are Free. /agent, /auto, and /runbook save-last <N> are Pro; a Free user who types them gets an upgrade prompt instead of a run.