9lives 0.5.0

SDK

• package.json declares the engine protocol, capabilities and prove fault
  kinds the build implements under "9l", so 9l doctor can compare an
  installed SDK with the CLI without loading it. A test keeps the field equal
  to the constants the bridge and prove channel send.

CLI

• 9l run --retries <n> passes Playwright's own per-test retries to every
  job, replacing the project's setting, so --retries 0 --attempts N lets 9l
  own all retries and their evidence. The --attempts help says that
  Playwright's configured retries still apply inside each attempt. Receipts
  record flakyTests, the tests that failed and then passed on a Playwright
  retry inside the attempt, with or without --sdk, and 9l run / 9l result print N test(s) passed only on a Playwright retry inside this attempt under the job (additive).
• With --sdk, receipts record reportedAssertions, the assertion steps the
  worker reported as passed in each executed test's last completed attempt,
  and passedWithoutAssertions, the tests that passed while reporting none.
  9l run and 9l result print N passed test(s) reported no assertion step under such a job, so a green test that asserted nothing is visible.
  These are the worker's report, not verified coverage; assertionCoverage
  stays unknown. Without --sdk the counts are absent, because Playwright's
  JSON reporter serializes only test.step steps. Receipt, run-result and
  run-status contracts gain the two optional fields (additive).
• Receipts record keepAttachments when --keep-attachments was given,
  failureAttachments for the attachment files the listed failures
  reference and retainedAttachments for those copied into the receipt
  directory. A request that found nothing to copy prints --keep-attachments: the listed failures referenced no attachment files under the job instead
  of silently doing nothing (additive).
• Each failed test in a receipt carries a class: assertion, timeout,
  action-timeout, environment (a connection, name-resolution or
  certificate error on navigation, by Chromium or Firefox error code),
  unexpected-pass or other. It comes from Playwright's structured status
  and its own error prefixes, never from message text, so a test quoting
  those words is still other. 9l run and 9l result show it after the
  failing line, so a target that never came up is not read as a broken test
  (additive).
• 9l scan --format json adds scoring to the result: the 0–100 scale, the
  points each severity costs (35 critical, 20 high, 10 medium, 4 low, 0 info),
  the grade thresholds (A 90, B 75, C 60, D 40) and the grade. The Markdown
  report prints the same rule under the score, and docs/cli.md and the
  scan_url description state it, so a score of 0 on a page with ten
  medium findings can be checked rather than guessed at.
• 9l scan reports the page's own aborted prefetch and RSC requests as one
  info finding per page, with occurrences and up to five distinct urls
  (omittedUrls counts the rest), instead of one finding per URL. Other
  failed requests stay one finding per URL. The grouped finding has no url,
  so its id no longer depends on which prefetch was aborted first; against a
  baseline from an earlier release the per-URL findings show as fixed and the
  grouped one as new, once.
• 9l doctor judges the SDK pair by the protocol and capabilities the
  installed @9l/playwright declares, the contract run --sdk checks in the
  handshake, instead of by version string. A different release that this CLI
  admits is reported as admitted but unqualified, with the install command
  for the matching release; a different protocol or capability set is
  incompatible and, under --sdk, a missing prerequisite. Under --sdk, a
  new sdk/faults check lists the prove fault kinds the SDK applies. An SDK
  that predates the declaration is reported as such.
• 9l doctor no longer tells a Node release above the qualified 24 to
  install Node 24. Below 22 stays missing with the install step; any other
  supported release is a warning without one.
• 9l prove accepts --config <file> and --project <name> (repeatable),
  applied to the baseline and every fault run as 9l run applies them. A spec
  outside the nearest config's testDir no longer has to move to be proved.
• When 9l run --sdk, 9l prove or 9l confirm cannot run a spec, the
  reason names only the missing package: @9l/playwright, @playwright/test
  or a package.json that declares it. Previously it asked for both packages
  even when one was installed.
• 9l heal says when Tier 1 and verification were not attempted. In the
  session JSON, and in the MCP heal_test result, tier1 and verified are
  null when the session ended before that stage ran (the original passed,
  ran zero tests or failed an assertion, or no candidate was proposed);
  previously both read as a failed zero-test run. 9l heal accepts
  --format json (the default, the session JSON) and --format text (the
  state and reason, one line per run with not attempted, and the verified
  candidate's diff; no JSON). An unknown option is reported as such; only a
  Python-only option such as --framework is referred to 9lives heal.

Install

• CLI: curl -fsSL https://9lives.run/install.sh | sh (macOS, Linux), or download the archive for your platform below and verify it against SHA256SUMS as described in docs/install.md (Windows ZIPs included).
• SDK: npm install --save-dev @9l/playwright@0.5.0 @playwright/test@1.64.0. Matching CLI and SDK releases are the qualified pair; 9l doctor --sdk now reports whether an installed pair is admitted.

Qualified with @playwright/test 1.61.1–1.64.0 on Chromium, Node 24 (22 minimum), built with Go 1.27.3.

Full Changelog: v0.4.9...v0.5.0