awslabs/aidlc-workflows v2.10.1-preview.20261003.1

AI-DLC Workflow 2.10.1-preview.20261003.1

on GitHub

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

Merged commits since preview source 6ecaaac:

• fix(release): git init the Cursor project in the lifecycle jobs, as setup says (#1649)
• fix(kiro): name the v2 engine in Kiro recovery and give doctor each harness's own advice (#1630)
• fix(doctor): find the Copilot CLI config on Windows and match folders the way Copilot does (#1608)
• fix(copilot): run AI-DLC's own commands without an Allow click in VS Code, and stop doctor's false failure (#1570)
• fix(tests): run live drives on CI's models when the suite starts inside a Claude Code session (#1617)
• docs(graph): name --plan-approval among the typed settings flags (#1638)
• fix(plan-approval): keep a skip said at a report-only compose gate (#1559)
• fix(hooks): always mark a hook child's payload session as the payload's (#1637)
• fix(doctor): no plugin warning on hosts without a plugin list (#1618)
• fix(doctor): say bun is fine when this project's hooks are already running through it (#1619)
• fix(reverse-engineering): ask the person to approve, even when the stage is reported complete (#1621)
• test(t121): give the 32-fingerprint FIFO case the file's case budget (#1622)
• fix(reverse-engineering): skip .NET build outputs in the scan and the source boundary (#1616)
• fix(routing): offer bugfix for long bug descriptions, and ask before /aidlc-init picks a scope (#1610)
• refactor(kiro): read the Kiro tree layout from the installed tree, not the row name (#1572)
• ci: bound live shards and decouple Full Suite from preview releases (#1587)
• docs: add a facilitator guide and stop promising "never trapped" (#1604)
• fix(workspace-detection): an empty Cursor project is greenfield, not brownfield from AI-DLC's install.ts (#1603)
• fix(plan-approval-guard): let the person park and resume during Code Generation (#1607)
• fix(copilot): keep every directive whole in VS Code's terminal result (#1563)
• docs: guards fail toward the person, one owner per rule, team rules win; AIDA's direction lens judges by the tenets (#1599)
• chore(lint): raise the aidlc-lib.ts size cap to 1.5 MB (#1605)
• test(t147): keep the Kiro rebind case's drift when the host is slow (#1578)
• fix(copilot): a continue the hook cannot match still moves on, never a dead end (#1585)
• fix: count only what the person types as their turn, not a subagent's brief (#1567)
• docs: the human drives: put AI-DLC's tenets in AGENTS.md and AIDA's review contract (#1581)
• test(plan-approval): pin one approval for a stock install whose rules come as one separate part (#1579)
• fix(plan-approval): approve the code plan once, even after the chat compacts or the work pauses (#1561)
• test(windows): temporarily skip launcher helper repair cases (#1571)
• fix(engine): move on after an approval, and say complete only at the real end (#1564)
• fix(plan-approval): approve the code plan once, even when the rules come in parts (#1560)
• test(t149): create the rebind case's second intent from another session (#1565)
• fix: deliver stop-hook error directives once with the engine's exact message (#964)
• fix(copilot): a mistyped or replayed continue receipt returns the current step (#1562)
• fix(kiro): read delete_file's targetFile in both Kiro adapters (#1558)
• fix(composer): start a custom plan from classic's settings and Guard Policy off (#1554)
• fix: resolve a lone committed intent record without joining it (#1476)
• fix(log): say a reply is already recorded when a second answer reuses it (#1525)
• fix(config): keep existing .gitignore rules and name the ones that hide shared records (#1107)
• test(windows): keep launcher fixtures consistent for preview releases (#1539)
• fix(log): refuse a decision or answer whose value arrived split, and steer Kiro IDE agents off it (#1512)
• test(kiro): pin the Kiro CLI adapter recording the human turn in the project (#1532) (#1534)
• feat(scopes): default every scope except enterprise to Guard Policy off (#1531)
• fix(stage-gate): record Request Changes in the person's own words (#1524)
• fix(tests): give the animated menu fixture a unique frame every tick (#1528)
• fix(tests): read macOS process identity from the process table ps reads (#1527)
• feat: support transaction locking on mounts without hard links (#1108)
• fix: read the person's reply in their own words at every question (#1513)
• test(engine): pin that a checkpoint Request Changes redoes a stage whose gate was approved (#1523)
• fix(kiro-ide): run a native install's /aidlc chat commands through the binary's engine routes (#1522)
• fix(config): wrap the setup check and setup-complete screens to the terminal width (#1515)
• fix(engine): accept a unit-major skip of the stage the walk is running (#1499)
• fix(config): check the transaction lock before first-run questions and name the storage fix (#1507)
• fix(stop-hook): let the turn end after next refuses a typed level word (#1514)
• fix(summary-confirmation): refuse the answer that could never count, and name the command that does (#1510)
• fix(composer): write the grid proposal inside the project, not the temp dir (#1496)
• docs: how to edit by hand and continue, and what runs in order versus in parallel (#1509)
• fix(code-generation): say the review runs after plan approval and generation (#1500)
• fix(config): refuse a mid-workflow pin that would split engine and hooks, and name the right recovery command on Copilot (#1511)
• fix(kiro-ide): run no terminal command while an approval gate waits, and record how refusals reach the model (#1497)
• docs(kiro-ide): explain supervised and autopilot, and the terminal trust prompt (#1495)
• fix(config): wrap first-run setup menus to the terminal width (#1502)
• fix(release): test and publish the commit a preview started on (#1508)
• test(stop-hook): pin that a summary Request changes still waits for its feedback (#1506)
• fix(copilot): fold the drive letter into the project identity so VS Code on Windows can continue (#1504)
• chore(plan-approval): retire the agent-driven approval commands the engine replaced (#1501)
• feat(scopes): let a scope or a person turn Plan Approval off (#1494)
• fix(stop-hook): close a logged decision on its checkpoint's own answer (#1466) (#1503)
• fix(guard): never block the answer to the engine's own question (#1486)
• feat(composer): keep a composed plan with its work, save it on request (#1458)
• fix(source): compare old source evidence under today's exclusions, and keep them out of the swarm commit (#1493)
• fix(workspace-detection): name every nested repo, not just the one with code (#1498)
• fix(kiro-ide): start a chat's session from its first prompt (#1489)
• feat(plan-approval): the engine asks for plan approval and keeps the fingerprint itself (#1490)
• fix(cursor): say when a project is outside a git repository (#1484)
• fix: stop workspace navigation from resuming workflows (#1121)
• fix: exclude incidental OS/editor artifacts from the source-floor walk (#1099) (#1235)
• docs(knowledge): guide coverage-ignore pragmas toward provider detection and fail-loud (#1388)
• fix(kiro-ide): guide first run through trust, reload, and the aidlc agent (#1482)
• fix(guards): make summary confirmation off the person's own switch (#1481)
• fix: Exclude tool-managed byproduct files from the source fingerprint (#1242)
• fix(windows): read PowerShell planning commands as PowerShell and keep aidlc.cmd arguments intact (#1474)
• docs: add a guided onboarding walkthrough for first-time V2 teams (#1453)
• fix: end the Plan Approval section at a sibling heading, not at any heading (#1244)
• fix(plan-approval): offer the plan-approval switch only where it lets the work through (#1479)
• fix(worktree): let Bolt worktrees work in repositories with submodules (#1483)
• fix(plan-approval): check the session before recording accepted drift (#1485)
• docs: use the engine worktree command in workshop mode and say which gate counts are for a new project (#1480)
• fix: route every engine ask through typed commands and start work once (#1426)
• fix(settings): honour a recorded bypass at every guard that reads it (#1478)
• fix(plan-approval): name the wrong heading, placeholder session, and contract defect in refusals (#1473)
• fix(tests): answer a menu only once its rows hold still (#1475)
• fix(config): always ask what to change when a config section is named (#1477)
• fix(native): keep Kiro IDE hook state in the project, not the installed runtime (#1472)

Source commit: 8cce765

Nightly test report

Full Suite failed for 8cce765cec3e in run 37094975120. 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
• t-exec-codex-compose-inflight.serial (1 failed assertions) in live-codex-1-Windows
  • t-exec-codex-compose-inflight - in-flight recompose over exec + exec resume > beat 1 stops at the gate with nothing applied; beat 2 resume-approves the SKIP flips
• t-exec-codex-journey-workspace.serial (1 failed assertions) in live-codex-1-Windows
  • t-exec-codex-journey-workspace (live codex-exec multi-repo·intent·space journey) > one feature spanning two repos, a 2nd intent, a non-default space — composed live over codex exec
• t-exec-codex-memory-include.serial (1 failed assertions) in live-codex-1-Windows
  • t-exec-codex-memory-include — Codex resolves the relocated method tree via @-mention (closes t156 test 8) > @aidlc/spaces/default/memory/org.md is reachable and its content loads into context
• t-exec-codex-status.serial (2 failed assertions) in live-codex-1-Windows
  • Codex fixture failure and retirement contract > Windows runner-owned fixture deletion is handed to post-job cleanup with a receipt
  • t-exec-codex-status — $aidlc --status on the shipped dist/codex via codex exec > no-state: status renders 'no active workflow' and scaffolds nothing
• t-tui-t73-intent-capture.serial (1 failed assertions) in live-claude-tui-3-Windows

Runner errors

• live-claude-tui-3-Windows: INCOMPLETE: required selected coverage was not fully exercised

Failed jobs

• Linux / result (1): Linux / result
• Windows / claude-tui (1): Windows / claude-tui 3/3
• Windows / codex (1): Windows / codex 1/1

Full logs are in the run 37094975120 full-suite-* artifacts.