github awslabs/aidlc-workflows v2.10.1-preview.20261003.2
AI-DLC Workflow 2.10.1-preview.20261003.2

pre-release2 hours ago

Warning: Full Suite failed for this source. A Full Suite failure report ends these notes.

Merged commits since preview source 8cce765:

  • fix(document-input): onboard a PDF or Word file the person named, without asking them to (#1720)
  • test: wait for a rendezvous file's whole line in t330 and t244, as t333 does (#1722)
  • fix(orchestrate): a setting typed with new work goes with the work the person picks (#1716)
  • fix(tests): refuse to build the coverage registry from a stale dist/ (#1715)
  • test(t250): cite the issue number alone in the Windows path comment (#1691)
  • fix(copilot): a copied attempt id is refused for what it is, with the command to run instead (#1701)
  • fix(copilot): a cd to the project before an AI-DLC command runs like the bare command in PowerShell (#1700)
  • fix(hooks): hook status files are written without following links (#1718)
  • fix(hooks): a question already asked is never asked again after the end-of-turn reminder (#1707)
  • test(e2e): add default-off opt-in to run native-root tiers on an overlay-FS host (#1556)
  • fix(engine): say which write put the build step out of date, in doctor (#1668)
  • test: split the three slowest test files so the merge queue finishes sooner (#1714)
  • fix(plan-approval): a lowered fence never refuses the build because its audit row could not be written (#1710)
  • fix(onboarding): say the hooks use the terminal's PATH, not ~/.bashrc (#1650)
  • fix(doctor): never run VS Code's stand-in copilot, so no install question shows (#1706)
  • fix(document-input): take a pasted document and a named file as the person gave them (#1696)
  • feat(copilot): say in the chat when AI-DLC's hooks have never run (#1662)
  • feat(workspace): say whether work is a new project or existing code, and ask when an empty folder gains code (#1655)
  • fix(cursor): never throw away the person's prompt; say once which work this chat is on (#1683)
  • ci: the merge queue retries a flaky test file once, and says so (#1712)
  • test(sdk-drive): keep the answer stream open while subagents still run (#1711)
  • fix(review): record a review whose only slip is a ## heading, with no second review (#1709)
  • fix(config): record or clear a bypass while a workflow is running (#1644)
  • fix(hooks): the Stop hook's normal waits are trace, not drops; doctor names real failures (#1689)
  • fix(engine): join the chat that migrates a flat project when its tool cannot name it (#1697)
  • fix(audit): keep one audit file per clone, and read a copied audit file once (#1664)
  • fix(documents): compare file identity exactly on Windows (#1703)
  • feat(hooks): an opt-in phase trace for hook processes, on in the Windows live legs (#1692)
  • fix(routing): do not ask about new work for an answer to the stage's open question or the ask's own option (#1520) (#1521)
  • ci: rebalance test shards and start the longest integration files first (#1699)
  • test(coverage): commit only per-unit entries so queued PRs stop colliding on the registry (#1705)
  • fix(cli): say what uninstall and prune remove, then do it, with no y/N question (#1695)
  • test: fix three test races that dropped PRs and reddened local runs (#1702)
  • fix(config): refuse adding a harness on another version under a running workflow (#1629)
  • fix(construction): keep finished Units' receipts when you switch to unit-by-unit or turn checkpoints on (#1654)
  • fix(engine): skip, add, jump, and change scope do what you named (#1694)
  • fix(plan-approval): every refusal names the step that ends it, under every Guard Policy (#1660)
  • fix(intent): archive and unarchive do what you asked (#1688)
  • fix(config): stop recording project-owned harness files as framework-owned (#1526)
  • fix(questions): take what the person already said instead of asking again (#1681)
  • fix(presence): never ask the person to reply again; say what happened to their reply (#1684)
  • fix(config): name the host session as the model lever on Copilot, Cursor, and Kiro IDE, and say how to choose model and effort (#1645)
  • fix(guards): a question never switches plan approval off (#1677)
  • fix(windows): replace the old launcher helper under an 8.3 install path, and say why when it cannot (#1666)
  • fix(copilot): end the turn after /aidlc intent or space selects another intent (#1583)
  • fix(code-generation): pick up an interrupted build at the first unticked step (#1586)
  • fix(scopes): bugfix skips the learnings question and summary confirmation (#1667)
  • fix(copilot): after a Unit's work is recorded, Stop tells the agent to run next (#1671)
  • feat(config): the setup check names Copilot's real trust switches and how to trust the folder for the CLI (#1623)
  • fix(engine): the work picker lists only work you can carry on (#1682)
  • test(change-control): wait for a barrier file's whole line, not just its path (#1680)
  • fix(engine): let a finished intent on a retired scope stop blocking next (#1632)
  • fix(composer): drop Kiro's scope write grant, refuse path selectors (#1640)
  • fix(agents): leave the host's own agents in the harness agents dir alone (#1544)
  • feat(ci): tell the author when a runner failure dropped their PR from the queue (#1678)
  • fix(lifecycle): wait out a busy machine lock when reserving a pinned release (#1625)
  • fix(plugin): count a Claude local or project install only in its own project (#1627)
  • fix(claude): keep generated runners explicit-only, as on Cursor (#1628)
  • fix(tests): find the e2e report from an isolated-file attempt directory (#1676)
  • fix(learnings): accept the candidate id surface emits in a selection (#1269)
  • docs: key the story-map rows the way traceability enumerates them (#1257)
  • fix(claim-sources): recognise contract headings behind a leading emoji (#1488)
  • fix(usage): price a point release on its own row, not its major (#1577) (#1601)
  • fix(engine): start new work over a finished workflow with the typed scope (#1535) (#1555)
  • test(t161): judge the owner's own release, not the waiter's new lock (#1672)
  • fix(ci): a timed-out evidence upload no longer drops a PR whose tests passed (#1651)
  • test(codex): use AWS default credential chain when AIDLC_CODEX_AWS_PROFILE is unset (#1536)

Source commit: 028b42c

Nightly test report

Full Suite failed for 028b42cd99c0 in run 37157174895. The preview build does not wait for these tests.

Failed legs: deterministic (skipped), production_guards (skipped), live_windows (failure).

Failing tests

  • t-exec-codex-compose-front.serial (1 failed assertions) in live-codex-1-Windows
    • t-exec-codex-compose-front - interactive compose over exec + exec resume > beat 1 stops at the gate with nothing written; beat 2 resume-approves and creates the intent

Failed jobs

Full logs are in the run 37157174895 full-suite-* artifacts.

Don't miss a new aidlc-workflows release

NewReleases is sending notifications on new releases.