[vibhavgo]

Summary

Surfaces the real outcome of the lerna publish steps (tee'd logs + ::warning:: annotations keyed on outcome, which continue-on-error does not mask) and uses what lerna actually published to trim Verify Publish's propagation wait — a storm-aborted wave publishes only a fraction of the tree, and waiting 15 minutes for versions that never existed was pure loss. Also caps per-package conflict polls at ~30s on mass-stall days so the settle pass cannot run packages × 5 minutes against the job timeout.

Linear: WCI-2113

Changes

• publish.yml: both lerna steps get id: + shell: bash (opts into -eo pipefail — the default run shell has no pipefail, and a cmd | tee pipeline without it masks lerna's failure) + 2>&1 | tee /tmp/lerna-{siblings,bitgo}-publish.log; two ::warning:: steps keyed on steps.<id>.outcome == 'failure'; log paths passed to Verify; timeout-floor comment with empirical evidence (healthy 43 min, storm 78 min — do not trim timeout-minutes: 100); EUNCOMMIT on the bitgo step documented as the desirable storm-day path (document-don't-fix: fixing it would run a 20-min prepack against missing dependencies; the healthy-day Nov 3 run proves the step publishes bitgo fine).
• verify-release.ts: parseLernaPublishedPackages + parseLernaAttemptedPackages (version-anchored regexes — excludes lerna's published N packages summary lines; lerna verb publish <name> marks every dispatched package, verified against lerna 10.0.1 output: 114 unique names in a healthy log = exactly the candidates), readLernaPublishSets (unions both logs into published + attempted sets with published ⊆ attempted; fail-open: unreadable logs or readable logs with no recognized lines — format drift — revert to waiting for every package with a one-line warning), shouldShortPollConflict (mass-invisible OR aborted-wave trigger, keyed on how many packages lerna reached), VerifyPackageOptions gains sleep/now/conflictPollTimeoutMs passthrough into publishWithRecovery, and verify() builds its wait-list from the attempted set — errored packages (TLOG 409 usually means the write landed anyway) get the wait instead of a needless bump plus dependent cascades; only never-dispatched packages skip it — and threads the 30s cap on mass-stall days.
• Tests: 12 new (59 total passing) — parse (scoped/unscoped/summary/emoji/WARN exclusion), fail-open ×3 (missing file, unreadable, no publish lines), short-poll truth table incl. boundaries, poll-cap passthrough (asserts 1 visibility + 3 poll checks vs the default 11).

Test Plan

• npx mocha --require tsx ./scripts/tests/generate-bitgo-shrinkwrap.test.ts ./scripts/tests/verify-release.test.ts — 57 passing; tsc --noEmit clean on changed files; workflow YAML validated.
• Two adversarial subagent review passes over the full diff: both "correct", first pass's two P3s (summary-line capture, comment overstatement) fixed and re-verified in the second pass against live files.
• Live: healthy-day dispatch unchanged (~43 min); storm-day dispatch expected ~63 min (from 78) with the ::warning:: visible on the run summary and the "Lerna published N of M packages" line in Verify's log confirming the trimmed wait.

[linear-code]

WCI-2113

[vibhavgo]

@claude review always

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

[vibhavgo]

@claude review

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}