qawolf/cli v1.9.0

on GitHub

Minor Changes

• 8c97c28: Update @qawolf/api-contracts to 0.25.0. Environment responses now identify
  the default environment, and qawolf run get includes attempt artifacts and
  failure diagnoses when available.

• d2460a7: Add qawolf runner screenshot, act, exec and keepalive, so a caller with a shell and its own vision model can close the see-and-act loop against an interactive runner.

  qawolf runner screenshot writes the runner's screen to a file, decoding the image on the way, and keeps the three ways it can have no image apart: the runner has not run anything yet, so nothing has started its screen (run a flow); the screen is up but cannot serve this instant (retry); the runner has no screen at all (launch a different image). qawolf runner act <action> performs one raw action in the computer-use vocabulary a vision model emits, validated against the published schema before it is sent, and accepts a whole action as JSON on stdin so a model's tool call can be forwarded unchanged. qawolf runner exec <file|-> evaluates a snippet against the live page, with --file to give it the scope of one of your own files; it reports whether the snippet ran, and points at the console stream for anything it printed. qawolf runner keepalive resets a runner's inactivity clock for a caller that pauses between actions.

• 7275358: Add qawolf runner, a command group for driving an interactive runner on the QA Wolf platform.

  qawolf runner launch starts one and makes it this directory's default; qawolf runner stop stops it. qawolf runner run <file> ships the current directory's runnable files and runs a flow on the runner, and with --follow streams the run's logs until the run settles. qawolf runner events <stream> prints a runner's journal one entry per line, so --tail, --since, --run and --follow compose with grep and jq; --json prints the whole envelope. Every --follow is bounded by --timeout (an hour by default), because reading a runner keeps it alive and billing.

  Runner-targeting commands take an optional --runner <id>, falling back to QAWOLF_RUNNER_ID and then to the runner stored for the directory. A command with no runner available launches one and says so, naming it, so a caller knows the browser it is now driving is fresh. A launch that fails still names the runner it was launching, because a launch whose answer was lost may have started one: relaunching that same id attaches to it rather than starting a second.

  Under --json, and so in CI, a followed run's logs are printed as journal entries rather than as bare text, which keeps everything the command writes to stdout parseable as one stream.

Patch Changes

• a76fbcd: qawolf runner screenshot asks for a runner that already has a screen instead of starting one, and qawolf runner exec names the missing page rather than the runner's image.

  A runner's virtual desktop starts with its first run, so a runner launched to serve a screenshot would have no screen and could only answer screen-needs-a-run. screenshot refuses instead, and names both things it needs: a runner, and a run on it. That also keeps the group readable, because no read command starts a runner.

  A node20WithPlaywright runner with no live page does run a browser and simply has nothing open on it yet, so qawolf runner exec points at the page rather than at the runner's image. Checking the image stays as the secondary possibility, since it is the case that never clears.

• 7275358: qawolf runner now passes on the reason QA Wolf gave for refusing a request, instead of only the status code.

  Running a file that is not a flow used to answer "runner.runFlow request failed (HTTP 400)" and stop there, while the server had already said which file was wrong and that an entry point has to be a flow file under src/flows. That sentence now reaches the caller, on run, stop, screenshot, act and exec alike.