Engineer mode
Re-run AI generation on an existing capture without re-browsing.
Engineer mode skips the capture step. It takes an existing run (HAR plus metadata) and re-runs the AI generation pass. Useful when you want to try a different model, prompt, or output language without re-doing the browser work.
It needs the run's HAR, so it works on runs from the auto and
agent-browser providers and from manual mode. Runs from the default
chrome-mcp provider save no HAR; engineer stops right away for them
(error_kind: misuse, exit code 2), and you run agent again instead.
When to use it
- The first generation missed something. Try Opus 4.6, or tweak the prompt.
- You want the same capture exported to a different language (Python to TS).
- You want to iterate on the generated client without paying the cost of re-capturing the site.
How to run it
Interactive:
reverse-api-engineer
# switch to engineer mode with Shift+TabEngineer mode works on your latest run with a HAR. The line above the input
names it, with its id and prompt. Type what should change, such as
add pagination, and rae regenerates that run's client with it. /engineer
lists the runs that have a HAR to pick another one, and a run id typed on its
own regenerates that run as it was.
The run's cost in /history includes every re-engineer of it, and the client
card shows what the last one cost.
Direct:
reverse-api-engineer engineer <run_id>From another agent or CI wrapper:
reverse-api-engineer engineer <run_id> \
--prompt "add pagination support" \
--json-streamUse --json when you only need the final result. Both JSON modes suppress
follow-up prompts and keep machine-readable output on stdout.
Find the run_id of a previous run with:
reverse-api-engineer listOr --json for a machine-parseable list:
reverse-api-engineer list --json | jqWhat gets regenerated
Everything the AI emits gets rebuilt: api_client.*, example_usage.*,
README.md. The HAR itself is unchanged. Engineer mode writes back to the
same run's scripts directory, so use version control or copy the output first
if you want to compare iterations side by side.