awslabs/aidlc-workflows v2.10.1-preview.20261006.2

AI-DLC Workflow 2.10.1-preview.20261006.2

on GitHub

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

Warning: Updating from the last release to this preview failed in its checks. Stay on your release or wait for the next preview.

Merged commits since preview source c86f6d9:

• fix(runtime): Codex reads its own install in a project shared with another tool (#2000)
• fix(code generation): an approved plan that changes before the build is named, and the person can go back to it (#1998)
• fix(hooks): say why tools are refused in a folder holding the AI-DLC install (#1848)
• fix(composer): the plan composer and reviewers read and write with their file tools (#1997)
• fix(questions): the summary the person confirms is on screen with its question (#1996)
• fix(resume): the pick-up line says where the work picks up (#1995)
• test(guard matrix): what a person meets when files change under approved work (#1867)
• fix(opencode): a copy's settings change never claims the team's opencode.json entries (#1992)
• fix(orchestrate): new work starts beside open work on a scope this install no longer knows (#1993)
• test(kiro): the contractions case reads the call through sh only where sh is the shell (#1991)
• fix(stop-hook): say in one line when AI-DLC carries on by itself (#1953)
• fix(construction): one question approves the late stage approvals when Units are built one at a time (#1981)
• fix(code generation): an interrupted build of an edited plan picks up where it got to (#1987)
• fix(routing): carry on with work in progress continues it without the routing question (#1977)
• fix(construction): the change line still reaches the person when one next settles the bookkeeping gates (#1990)
• fix(plan approval): a setting typed with the reply to the plan question does both, in the person's own words too (#1886)
• fix(opencode): setup and refresh keep the team's own opencode.json (#1937)
• fix(pick): picking work up in a clone carries on in one answer (#1936)
• fix(statusline): a Unit walk counts Units, not whole-stage checkboxes (#1986)
• fix(config): config --show shows every section without changing anything (#1983)
• fix(composer): a short plan offer, with the stage table and scores on request (#1960)
• fix(guard-policy): a hand edit of a reviewed document is said as the stage's documents, not another document's name (#1984)
• fix(construction): under strict, a teammate clone of an approved Unit gets a fresh review instead of a refusal (#1982)
• docs(reference): the conductor reads the person's words, there is no resume menu, and status shows no receipt-less history (#1980)
• fix(replies): an apostrophe in "it's" or "users' files" is a letter, not a quote (#1935)
• fix(package): each tool ships only its own protocol bindings (#1979)
• fix(context): a resume no longer reads the engine-owned format docs (#1978)
• fix(guard): the question that offers ways on names the stage in plain words (#1976)
• fix(resume): a stage resumed with blank answers hands back what the person already replied (#1930)
• fix(review-freeze): the refusal names next, and next asks the recovery question once (#1969)
• fix(kiro): the Review brief prints while a gate waits, and the presence floor says one line the person can read (#1975)
• fix(guard): Request Changes on a recovery question names the report to run (#1974)
• fix(codex): commands keep their own words once Codex has given a tool the session (#1972)
• fix(code generation): a picked-up build says only what is certain, and both lines count the plan the same way (#1967)
• test(construction): on Windows an approved Unit stays approved while another program holds one of its files open (#1966)
• ci(merge-queue): a selected macOS file keeps its affinity group and the shard rules (#1965)
• fix(settings): a setting typed with the request reads set by you on the work it creates (#1962)
• fix(workspace): what the folder is needs the person's word after the question, not the /aidlc that led to it (#1961)
• test(person turns): an unbacked decision keeps the turn ledger and the audit trail (#1958)
• fix(skill): "from here on, build one unit at a time" is carried out at an open question (#1955)
• fix(review): an empty placeholder row in a review's findings is not a finding (#1963)
• fix(re-entry): "/aidlc take me back to X" is that jump, and a bare /aidlc after a park carries on (#1959)
• fix(construction): one next after the last Unit approval records the bookkeeping stage gates (#1950)
• fix(kiro-ide): a shell command that changed nothing no longer loads the engine for its two shell hooks (#1946)
• fix(install): a setup that has put every file in place is never undone by a busy scratch folder (#1947)
• test(t244): a Windows release fixture server that stalls before it listens starts once more within its own wait (#1945)
• fix(next): words typed with a different --scope over open work ask once, new work first, never dropped (#1939)
• feat(engine): one read-only clock, aidlc engine now, for every timestamp (#1942)
• docs(guide): pre-work guides for a vision note and a technical environment note (#1941)
• test(runner): the merge queue's one retry also covers a file whose only failures are case timeouts (#1920)
• fix(park): a park the run cannot take says why in words (#1943)
• fix(status): a tailored plan's settings read as the approved plan's (#1940)
• fix(kiro-ide): an approval that waits names the same step doctor names (#1925)
• fix(statusline): the line names the step the person is on in a Unit walk (#1938)
• fix(opencode): a request the Windows launcher cannot carry reaches next through a file (#1904)
• fix(stop-hook): the stop note is one line the person can read, naming the next step (#1931)
• fix(plan): a tailored plan is named as the gate showed it, not after its base scope (#1934)
• fix(plan offer): each plan the offer lists carries the engine's own stage count (#1933)
• fix(routing): the new-work question counts only the work that is in this checkout (#1932)
• docs(guard-policy): a switch typed before any work exists is described the same everywhere (#1888)
• fix(swarm): with Guard Policy off or relaxed, a Unit changed after its review keeps the review (#1910)
• fix(change lines): a Unit's change lines name the Unit in plain words (#1892)
• fix(config): AI-DLC's read-only commands and turning a check back on run without a prompt (#1917)
• fix(checkpoint): a Unit approval never sends the agent looking for a session id (#1890)
• fix(deps): pin proxy-addr to 2.0.8 for the critical IP-spoofing advisory (#1915)
• fix(construction): a re-check of an approved Unit's documents reviews the right stage, one stage at a time (#1885)
• fix(review): the person's approval goes over a review that never finished (#1897)
• fix(kiro-ide): an approved plan that changed since builds on under a lowered Guard Policy (#1896)
• fix(jump): a backward jump tells the person how to return (#1901)
• fix(plan approval): what the person asks for while a code plan waits runs, and the question stays open (#1874)
• test(copilot): the hook-health chat joins its new work from the host's event after the command (#1912)
• fix(verification-command): the project check question never sends the agent looking for a session id (#1898)
• fix(kiro): a request with a line break or a backtick reaches the engine whole, quoted (#1899)
• fix(replies): "approve, and turn plan approval off" does both, and a stop under autonomy parks (#1887)
• fix(dispatcher): config set collaborators changes the setting (#1907)
• fix(guard): a Unit done without its receipts gets a question the engine can send (#1973)
• fix(replies): plan approval off typed beside --skip or --project-type is the person's too (#1902)
• fix(hooks-off): the first request carries on after the restart the step asks for (#1900)
• fix(stop-hook): read the Codex exec_command shell call so stopped work carries on (#1970)
• fix(kiro-ide): AI-DLC commands keep running with no card after a persona is dispatched (#1968)
• fix(plan-approval): I'll edit the files after an earlier answer opens edit mode (#1889)
• fix(learnings): no warning lines when the stage diary path is recomputed (#1905)
• fix(re-entry): a redo does the stage the person named, and "take me back and stop there" stops there (#1903)
• fix(plan approval): the refusal while a plan waits for approval is one plain sentence for the person (#1891)
• fix(kiro-ide): register review-freeze and state-transition-guard on Kiro IDE and Kiro CLI v3 (#1636)
• fix(next): a utility flag inside the person's request is part of it (#1954)
• fix(guard): under Guard Policy off, no check says it stood aside on every write (#1964)
• fix(construction): a redo the person asked for of an approved Unit is asked about once, as new work (#1949)
• fix(construction): every tool gets the same Construction instructions (#1957)
• fix(construction): a review recorded while the source could not be read keeps its Unit approved once the source reads again (#1956)
• fix(config): a copy's settings change keeps one AI-DLC part in AGENTS.md and .gitignore (#1951)
• perf(sensors): stop per-file tsc/eslint during code generation (RFC #1911 quick win) (#1944)
• fix(construction): a Unit restored as approved on a fresh clone can be approved again (#1948)
• ci(merge-queue): macOS unit shards run the files that name macOS and the ones the change touches (#1929)
• fix(construction): a Unit done but never recorded is told its receipt step, not the stage again (#1922)
• fix(construction): an approved Unit keeps its approval when its review evidence cannot be checked here (#1923)
• fix(construction): an approved Unit stays approved on a fresh clone (#1928)
• fix(construction): a redone Unit step has its own stale-review recovery (#1927)
• fix(gates): once the approval question is put to the person, only their reply to it records a rejection (#1924)
• fix(review brief): a Unit's own approval shows only that Unit's review (#1919)
• fix(construction): a deleted design document is made again and reviewed under any review cap (#1918)
• fix(validity): under relaxed and off an edit to a finished stage's document is not raised as out of date (#1916)
• fix(construction): under relaxed and off a wave entry edited after its completion stays settled (#1914)
• fix(construction): with reviews off, an edit between "Approve Unit?" and the answer is said once, not asked again (#1913)
• fix(prose): the person's words on a command line are single-quoted everywhere (#1875)
• fix(questions): the own-words tip names the free-text row as each tool shows it (#1893)
• fix(construction): a document changed before its completion and then put back opens the checkpoint (#1909)
• test(coverage): CI builds the coverage registry fresh and fails only if coverage drops (#1908)
• fix(kiro-ide): AI-DLC's allowed commands run without an approval card again (#1895)

Source commit: f228a77

Update from the last release

aidlc update from the last release to this preview failed in the preview run.

Nightly test report

Full Suite failed for f228a77266c6 in run 37536749291. The preview build does not wait for these tests.

Failed legs: scope_runs (failure), live_linux (failure), live_macos (failure), live_windows (failure).

Failing tests

• t-guard-live-chat-lowering.sdk (1 failed assertions) in live-claude-sdk-1-macOS, live-claude-sdk-2-Linux, live-claude-sdk-2-Windows
  • production guards: lowering from a live Claude chat > typed switches and plain words both lower the active intent's guards, as the person's
• t-guard-matrix-classic (1 failed assertions) in guard-matrix-macOS
  • classic as it ships, two Units > Unit 2 edits Unit 1's approved file without claiming it, Guard Policy off, review advisory, plan approval on, classic as it ships: nothing refuses or asks the person again, and the run reaches done
• t138-scope-exclusion-counts (1 failed assertions) in live-claude-sdk-2-Windows
  • t138 scope-exclusion counts (metamorphic invariant, sdk) > every SKIP-for-security-patch stage emits zero STAGE_STARTED (SKIP set derived from scope-grid.json)
• t238-user-stories-mob.sdk (1 failed assertions) in live-claude-sdk-2-Linux, live-claude-sdk-2-macOS
  • t238 user-stories mob topology (Claude SDK live) > three mutually blind supports write evidence, the lead integrates, and approval then succeeds

Failed jobs

• Linux / claude-sdk (1): Linux / claude-sdk 2/2
• Linux / result (1): Linux / result
• macOS / claude-sdk (2): macOS / claude-sdk 1/2, macOS / claude-sdk 2/2
• macOS / guard-matrix (1): macOS / guard-matrix / test
• Windows / claude-sdk (1): Windows / claude-sdk 2/2

Full logs are in the run 37536749291 full-suite-* artifacts.