qawolf/cli v1.30.0

on GitHub

Minor Changes

• 031c0a7: qawolf agent send and qawolf agent get are now hand-written commands with a --follow flag. With --follow, the command stays attached to the session. It prints each reply once, as it arrives. It exits when the session settles: 0 for completed, non-zero for failed or cancelled.

  A followed session can ask a question. In a terminal, the CLI shows the question and asks for an answer. It offers a list when the AI gives a fixed set of answers, and free text when it does not. It sends the answer to the same session and continues to follow. In a pipe, a CI job or agent mode, the CLI prints the question and the command that answers it, then exits 0.

  In --json and --agent mode, a follow ends with one line that holds the whole session object. This is the same object qawolf agent get answers with. A script can read the final status, the session id and the url from stdout.

  qawolf agent send remembers the session it starts. A later qawolf agent get --follow in the same directory needs no id. --session <id> and QAWOLF_SESSION_ID name a different session. --timeout sets how long a follow waits. The default is 30 minutes.

  qawolf agent send --file-paths <path...> names files the request is about, uploaded first with qawolf file requestUpload. The AI reads them from team storage instead of the message.

  Two flags changed name from the generated commands in 1.23.0. The message is now the first argument of agent send, not --message. --session-id is now --session on both commands. --environment-id still reads QAWOLF_ENVIRONMENT, and --workspace-id is still available for a credential that is not bound to one workspace.

  The QA Wolf platform reports only that a session is at work. A follow prints each reply as it arrives and shows a wait line between them. It shows more when the platform reports more.

• 15759eb: qawolf runner launch now says where the runner can be seen: a QA Wolf page showing its live screen, which a person can also drive with their own mouse and keyboard. A freshly launched runner has no screen yet — the page waits until the runner's first run starts one. It prints the address whether it started a runner or attached to one already running under that id, and so does a command that launched its own runner. --json carries it as url, and qawolf runner list --json carries it on every runner; the table leaves it out, since an address beside a 63-character id outgrows a terminal.

  The page opens for anyone on the runner's team, however the runner was launched, so the address is worth handing to a colleague who asks what your runner is doing.

Patch Changes

• 5121dba: qawolf flows pull, and the runs that pull on your behalf, now mirror team-storage assets with an organization or user API key. The pull used to stop with "Team storage requires a team API key" because such a key reaches many teams and names none. The environment being pulled names its team, so the pull now reads that team's storage. A team API key and a browser session behave as before.
• 36ea00d: Flows that import startWireGuard from @qawolf/testkit no longer fail with SyntaxError: The requested module '@qawolf/testkit' does not provide an export named 'startWireGuard'. The CLI now installs @qawolf/testkit 1.2.1, the version that exports the helper.
• a040bd1: The compiled binary runs web flows again. In 1.29.0, every qawolf flows run from the binary failed with Cannot find module '@napi-rs/keyring'. The worker subprocess that runs a flow loaded the keyring addon at startup, from a directory where it does not exist. The keyring now loads on first use, when a command reads or writes stored credentials, and @qawolf/flow-targets ships inside the bundle.