Documentation

Slash commands

Commands and keys of the interactive session.

Running reverse-api-engineer opens a full-screen session: the conversation scrolls above an input bar fixed to the bottom. Type / to open the command menu; Tab completes and Enter runs the highlighted command.

CommandWhat it does
/newEnd the open run and start a new capture.
/modelChoose the model.
/effortHow hard the agent thinks, for Claude models (low to max).
/modeSwitch between agent, manual and engineer.
/settingsSDK, agent provider, models, output and sync, changed in place.
/historyPast runs with their cost; open one to resume or re-engineer it.
/resume [run_id]Show a past run's conversation as it was and continue it, on any SDK.
/outputEverything from the session, unfolded.
/openOpen the latest generated client.
/clearClear the conversation and its spend.
/helpWhat rae does, its modes, commands and keys. Alias: /commands.
/exitQuit. Alias: /quit.

/resume works for runs that saved their SDK session, which agent and engineer runs do from this release on. Older runs cannot be resumed.

Keys

KeyWhat it does
EnterSend. While the agent works, the message is queued.
EscStop the run, or close a menu.
↑ / ↓Earlier prompts; in a menu, move, around from either end.
TabComplete a command.
Shift+TabCycle the mode: agent → manual → engineer.
Ctrl+ECycle the effort level.
Ctrl+OShow the full output of folded steps.
PageUp / PageDownScroll the conversation. End jumps back to the latest.
Ctrl+CStop the run, else clear the input, else quit.
Ctrl+DQuit.

Click a file or URL in the conversation to open it.

The status row under the input shows the run's id, which /resume and reverse-api-engineer show take. Once the agent has verified the client, its step lists the run's folder, and with real-time sync on, the synced copy in ./scripts, as full paths.

When the agent asks a question, its options show under the input: pick one with the arrow keys, press Space to pick several when the question allows it, or type your own answer.

Direct subcommands

Outside the session, you can use direct subcommands:

CommandWhat it does
agentRun an autonomous browser capture and generation pipeline.
engineerRe-run generation against an existing capture.
listList generated scripts and runs; supports --json.
showInspect one run; supports --json.
runExecute a generated client with its language toolchain.

In a terminal, agent and engineer open the session in their mode and start the run, keeping options such as --headless and --output-dir for the prompts you type next. They print plain output instead with --json, --json-stream or --no-interactive.

Scripted flags are command-specific: agent, engineer, and run support --json and --json-stream; list and show support --json; and run also supports --no-interactive for script selection and dependency prompts. See Scripted usage.

Manual mode is available only inside the interactive session because it opens a visible browser for human-driven capture.