v0.12.0

Added

• Environments from the CLI. project env list | create | update | delete | set-default <project-id> manage a project's environments over /api/cli/v1/projects/{id}/env with the existing read:projects / write:projects scopes; passwords come from --password-file and are never printed, and delete refuses the default environment (the server soft-deletes, so run history keeps naming it). --env <name> on test run (including --all), test rerun and test result --history targets or filters by environment; an unknown name is a server-side validation error that lists the project's environment names, never a silent fallback. project create and project update accept --local <port> (with --local-host and --skip-preflight) for an app on this machine — the CLI builds the loopback URL and probes the port first; a loopback --url is refused and pointed at --local, and a bind-all 0.0.0.0 / :: is refused everywhere. Every run now names an environment on the server (--target-url and --local match by origin or create one), and run, result and history payloads carry environment {id, name}; the history table gains an ENV column and authMode reports account / otp / manual / public. Requires a backend that serves the environment facade; against an older backend the project env verbs return not-found and --env is rejected.
• TESTSPRITE_CLIENT — let a wrapper identify itself. A tool driving the CLI (the GitHub Action sets github-action/<ref>) can set TESTSPRITE_CLIENT=<name>/<version>; the value is appended to the User-Agent on every API request and the telemetry beacon (testsprite-cli/<version> (<client>)) and reported as the telemetry client field. The value is validated against a tight <name>/<version> grammar; anything else is ignored silently and never echoed. Without the variable every request is byte-identical to before.
• CI context and outcome counts on the telemetry event. The one per-invocation event now carries ciProvider (github / gitlab / circleci / buildkite / other / none), and under GitHub Actions ciEvent and repoHash — a salted, truncated hash of the repository, never the name. test run --all, testlist run and test run <id> --wait add outcome counts (accepted / conflicts / deferred / skipped, passed / failed / blocked / timedOut) and conflictReason, the most frequent reason a case did not dispatch; ci init adds platform, force, workflowExisted and projectResolved. All fields are low-cardinality numbers and closed enums; the privacy contract (no URLs, ids, messages, argument values or repository names) is unchanged, and so are the opt-outs.
• Pinned-version warning in GitHub Actions. A ci init-generated workflow pins cli-version, so the job keeps running that release forever. When the backend reports that the running version has fallen below its minimum supported version, a job under GITHUB_ACTIONS=true now gets one ::warning title=TestSprite:: annotation naming the pinned version, the floor, and the fix (testsprite ci init github --force) — on stderr, so --output json on stdout stays parseable. The plain TTY advisory is unchanged off Actions. This keys off the version floor the backend already sends on every response; the npm "newer version available" check is (still) skipped in CI, so the warning does not fire for a pin that is merely behind the latest release.

Changed

• Batch billing refusals exit 12, like a single run. POST /tests/batch/run and testlist run may now report per-case insufficient_credits / billing_hold conflicts, and both are named in the conflict summary with the server's message. When nothing dispatched, nothing was rate-deferred, and every conflict is insufficient_credits, test run --all (with or without --wait) and testlist run exit 12 with the same INSUFFICIENT_CREDITS envelope and billing nextAction a single test run prints — previously a generic "already in flight" conflict, exit 6. A backend that answers that case with a 402 envelope directly is mapped identically. A batch refused entirely for a billing hold comes back from the current backend as the standard 403 FEATURE_GATED envelope (details.reason: 'billing_hold') and exits 13 with the server's nextAction — the same as a single test run; per-case billing_hold conflicts appear only on a mixed batch, and an older backend that folds every case into them still exits 6 with the hold named. Every other conflict mix keeps its current exit code.
• The CI summary counts never-dispatched tests as skipped, never as failed. The --summary-file JSON (on test run, batch test rerun, and testlist run) gains a skipped field, and failed now counts only dispatched runs that did not pass. Previously a test that never ran — rate-deferred, conflicted (already in flight), not found, or skipped by the engine — was counted under failed, so a batch could exit 0 while its own artifact claimed failed: 1 and the checks tab showed a red annotation: three surfaces telling two stories. The annotations split the same way — a dispatched run that failed or timed out still annotates as ::error::, while a never-dispatched test annotates as ::warning:: — and the job-summary headline reads N/M passed (F failed, S skipped, T timed out). Migration: a pipeline that parses the summary JSON and gates on failed > 0 no longer fails on conflict/not-found/deferred members; gate on the exit code (which already covers every case where nothing ran: deferred → 7, all-conflict → 6, all-not-found rerun → 4, zero-dispatch → 5), or on failed + skipped to keep the old behavior.
• test rerun --all fails (exit 5) when the resolved rerun set is empty, instead of a silent exit 0. No tests matching --filter/--status/--skip-terminal — or a project with no tests — used to print one stderr line and exit 0 with no CI artifact: a rerun gate that greens on zero runs. It now emits the summary + annotation and fails with exit 5, exactly like test run --all (0.8.0); pass the new --allow-empty to opt back into exit 0 when an empty sweep is expected.

Fixed

• A batch rerun where every id has no replayable run exits 4, not 0. test rerun a b --wait with both ids in notFound dispatched nothing, printed two red annotations and an artifact claiming two failures — and exited 0, so a CI gate passed on zero runs. It now exits 4 (NOT_FOUND, the same code testlist run uses for a --case miss) with a nextAction pointing at test run <id> for a first fresh run, on both the --wait and non---wait paths. A mixed batch is unchanged: notFound ids surface as skipped rows without failing an otherwise-passing batch.
• An empty frontend stepSummary renders as n/a, not 0/0. A frontend run card whose step summary carried no steps printed steps 0/0 (passed=0, failed=0), the same shape a backend run uses to say "not applicable"; it now prints n/a like the backend card. Community contribution (#336).
• A month-restricted cron schedule is no longer described as "daily". The human-readable schedule description ignored the month field, so 0 9 * 1 * read as "every day at 09:00"; the month restriction is now part of the description. Community contribution (#319).
• DOCUMENTATION.md manual-authentication commands match the current setup / auth surface. Community contribution (#342).