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 561a528:
- fix(engine): picking parked work up in a fresh clone carries it on, instead of asking you to resume again (#2664)
- fix(cursor): native setup leaves no AI-DLC file at your project root, and your own install.ts no longer stops it (#2662)
- fix(build-and-test): a requirement marked N/A with its reason is not reported unbuilt at the Build and Test gate (#2658)
- fix(learnings): one selections file per piece of work, so two chats never save each other's picks (#1592) (#2581)
- fix(linter): a TypeScript ESLint config (.ts, .mts, .cts) gets a real lint verdict instead of a quiet tool-unavailable pass (#2202 follow-up) (#2595)
- fix(hooks): a scope change or a check typed with next ends the turn when its step names your words, and Copilot reads those words as words, not a path (#2643)
- fix(engine): Code Generation completes only once the plan it builds from is approved (#2642)
- fix(switches): turning a check back on clears your own local switch for it and says so, with no question (#2636)
- fix(kiro): at an approval question, AI-DLC's own commands run; your approval still waits for your reply (#2638)
- ci: PR checks and the merge queue finish sooner: sixteen unit shards and the guard matrix in two jobs (#2640)
- test(e2e): the nightly drives one reply per question through Claude Code and reports what it misreads (#2628)
- fix(swarm): a side-by-side landing that clashes in a sibling repository finishes there too (#2635)
- test(e2e): the live answers check sends every reply it drives, and meets each question the way a fresh chat can (#2634)
- fix(engine): the approval gate shown again ends the turn, and an asked change at it is said with it (#2625)
- docs(codex): say that the shipped project config opens the sandbox network, and how to close it (#2633)
- ci: a pull request runs the installer ShellCheck that the preview and release gates run (#2632)
- fix(build-and-test): a requirement built through its stories is not reported uncovered at the Build and Test gate (#2630)
- fix(stages): the code scan lists with its own file tools and only reads, so it asks to allow no git command (#2631)
- fix(skills): turning plan approval or summary confirmation back on asks before clearing the setting that keeps it off (#2629)
- docs(agents): a change in AI-DLC itself reaches the person only when their work needs it (#2627)
- fix(config): setup lists the two Kiro harnesses as Kiro CLI (V2) and Kiro, and Enter takes Kiro CLI (V2) (#2622)
- docs(code): hook comments say only what the code cannot (#2626)
- fix(config): a refresh replaces an agent file Kiro upgraded when this release changes what it allows (#2605)
- docs(guide): the progress diagrams, two setting passages and the worked examples describe the lines AI-DLC prints now (#2624)
- test(release): the update check also carries a copy-channel project's open work through its own refresh from the last release (#2621)
- fix(review): the review block at a gate says only what changed since the last review, and no "why now" label (#2623)
- fix(kiro-ide): a lookup you ask for on your MCP server runs while the code plan is written, not only at its question (#2620)
- feat(stages): Requirements Analysis cites the intent statement and scope document where they exist (#2210) (#2618)
- docs(tests): test comments point by name, never by line, and the pointer lint reads tests/ (#2617)
- fix(engine): the plan offer says how big the plan is, and stops there (#2616)
- docs(stages): every stage that lists FR IDs says a requirement is traced through its sub-requirements (#2210) (#2615)
- fix(swarm): a side-by-side landing that clashes says where and what to do, and the agent finishes it (#2599)
- fix(engine): the progress line counts only the stages your plan runs, and a failing check waits for you at the gate (#2614)
- fix(engine): "skip Functional Design and approve the rest" at a question about several stages approves the rest (#2613)
- fix(guards): a review pass past the cap or a missing Reverse Engineering handoff is the agent's own step, not a question for the person (#2608)
- fix(switches): a check you turn back on while a setting keeps it off says so, and offers to clear it (#2607)
- fix(engine): the change you ask for in the message that brings up the code plan question is your answer to it (#2611)
- fix(install): the release checks pass install.sh again (shellcheck SC2012 on the root-home owner read) (#2606)
- fix(codex): a project copied from 2.11.0 refreshes to the next release with its own tool, and its config.toml stays the person's (#2604)
- fix(gate): a check that reads the whole stage runs once per output folder, so its finding is said once (#2210) (#2596)
- fix(validity): an AI-DLC update that changes a finished stage asks nobody to redo it (#2603)
- docs(code): comments point by name, never by line, and AGENTS.md says what a comment is for (#2602)
- fix(guards): the state tool alone refuses a direct stage-status command, in every spelling; Copilot asks only when it would land (#2601)
- fix(sensor): traceability counts sub-requirements for their requirement, asks a Unit for its own NFR rows, and says where an N/A row goes (#2210) (#2592)
- fix(engine): the step that starts a stage carries the lead's persona, so nothing has to be read before it (#2580)
- fix(kiro): after Kiro CLI's 3.0 upgrade the developer and the conductor write only what Kiro CLI 2 lets them write (#2598)
- fix(cursor): a crashed AI-DLC guard refuses nothing, as on every other harness (#2594)
- fix(sensor): upstream-coverage checks the inputs a stage requires, not the optional ones (#2210) (#2593)
- fix(plans): the plans the agent proposes from say which ones split the work into Units (#2591)
- docs(kiro): the guides name Kiro's own choice that approves a side-by-side build's workers for the rest of the chat (#2590)
- fix(switches): a check you turn off is off first, and its own line then offers the one wider place (#2583)
- fix(log): log review --help names --person-asked on the request line (#2585)
- docs(code): comments in the Bolt, swarm, worktree and learning tools say only what the code cannot (#2579)
- fix(package): a project on 2.8.0 to 2.11.0 refreshes to the next release without stopping (#2577)
- fix(hooks): /aidlc --help is answered by the prompt hook on Claude Code and Codex, with no command run first (#2576)
- fix(kiro): a side-by-side worker writes its unit's own records without asking (#2573)
- fix(engine): a rebuilt Code Generation Unit is told to request its review, also after a reopen (#2487) (#2511)
- fix(hooks): a later prompt leaves Completed and Archived work alone (#2562)
- test(live): a rule for what live tests assert, and a lint that holds them to it (#2570)
- test(release): the update from the last release carries the work a person left waiting at a gate, with nothing asked again (#2572)
- docs(tests): e2e, smoke and test-harness comments say only what the code cannot (#2566)
- fix(questions): a new questions file names who would answer each question and keeps each trade's questions together (#2559)
- fix(doctor): the merge line says a merge started over 60s ago has no finish recorded after it, not "within 60s" (#2567)
- fix(config): a preset you pick in first-run setup stays when you change the harness and change it back (#2565)
- feat(install): an opt-in --allow-root lets install.sh and aidlc unins… (#2482)
- fix(engine): --help on an engine subcommand prints its flags instead of an error and an audit row (#2449) (#2504)
- fix(construction): a verification command file named from the project folder is read, not looked for under the record twice (#2451) (#2503)
- fix(code-generation): a reopened Unit keeps its earlier source claims (#2490) (#2501)
- fix(sensors): the traceability sensor reads story-map cells in inline code (#2488) (#2497)
- docs(guide): say where aidlc/ lives and what carries over between spaces and projects (#2515)
- test(scope-runs): the Cursor and opencode trees run the same journey, so every harness tree has one (#2563)
- test(windows): fixture removal retries, the job list settles, bounds come from the call, and the CI weights are refreshed (#2557)
- docs(tests): integration test comments say only what the code cannot (#2560)
- test(knowledge): the oversized-file test runs on macOS in the merge queue again (#2561)
- fix(skill): the /aidlc skill keeps what only it carries (#1330) (#2549)
- fix(review): another review pass past the stage's passes runs when the person asked for one, not after any message (#2552)
- fix(swarm): a side-by-side batch runs the tests you approved through AI-DLC, without asking you again (#2550)
- test(live): a live file the model service failed says so, and Windows Codex sandbox setup gets a second run (#2554)
- fix(kiro): after Kiro CLI's 3.0 upgrade only AI-DLC's own commands run unasked, as on Kiro CLI 2 (#2556)
- test(scope-runs): the Kiro CLI and Kiro IDE trees run the same journey through their own hooks (#2555)
- fix(engine): a missing pipeline handoff or collaborator note names the agent's step, never the switch that turns the check off (#2553)
- docs(tests): unit test comments say only what the code cannot (#2548)
- fix: a request left in aidlc/.aidlc-request-text/ is git-ignored, and Kiro settings AI-DLC cannot parse read as unknown, not as Workflows off (#2539)
- fix(kiro-ide): AI-DLC no longer refuses an MCP call while the code plan waits for your approval (#2538)
- fix(kiro): AI-DLC's own commands stay allowed after Kiro CLI's 3.0 upgrade (one allow per tool script) (#2547)
- fix(copilot): on the Copilot CLI, a brief AI-DLC sends a subagent is no longer recorded as a message you typed (#2544)
- test(scope-runs): the Codex and Copilot CLI trees run a whole scope through their own hooks, with two Units, a fresh clone and a second chat (#2542)
- fix(engine): the plan question waits for the person's reply before work starts (#2536)
- fix(update): page the preview releases list so update works past 1 MiB (#2537)
- fix(config): an agent's config flags --bypass of a check needs the person asking for it (#2541)
- fix(next): the hooks that read a next line take its flags the engine's way, so a word of yours that looks like a flag is not taken as one (#2518)
- test(live): live tests check what the person meets and the record, not the route the agent took (#2540)
- fix(kiro): AI-DLC's specialists on Kiro use the MCP servers you add, each call asking first (#2535)
- docs(code): comments in the harness manifests and emitters say only what the code cannot (#2533)
- docs(code): comments in more core tools say only what the code cannot (#2531)
- fix(config): a refresh keeps the agent files Kiro upgraded for its 3.0 engine and brings in the rest (#2529)
- fix(stop-hook): a gate you answered is recorded before the turn ends (#2532)
- test(live): the golden journeys and the Codex journeys meet stock permission settings, and AI-DLC's own commands never prompt (#2530)
- fix(plan-approval): while a side-by-side batch plans hold, its commands run as a serial build does (#2528)
- fix(session): on a busy machine, SessionStart and every command find their chat, on every OS (#2527)
- docs(code): comments in the quiet core tools say only what the code cannot (#2526)
- fix(answers): a choice left to the agent is recorded only with words the person said (#2522)
- fix(review): honor registry exclusions in directory source claims (#2475)
- fix(kiro): on Kiro CLI, config models --show, doctor and the guide say agents start on your saved model, not a /model switch (#2405)
- docs(code): comments in scripts and the test-pro plugin say only what the code cannot (#2525)
- test(live): a live run that sees AI-DLC act for the person stays red; the retry no longer hides it (#2521)
- fix(install): an IDE open during the Windows install is told to quit and reopen, not reload (#2396) (#2397)
- fix(status): status counts the stages the person was shown (#2470)
- fix(learnings): a check you keep at a gate joins the stage's sensors and leaves its header whole (#2517)
- chore(tests): remove the August Kiro routing recording and the test that only re-read it (#2512)
- docs(guard-policy): under strict, a build of Units in parallel asks again for your own change after the plan was approved and stops the landing on it (#2516)
- fix(kiro): Kiro CLI's 3.0 engine dispatches each persona with its tools (#1394) (#2514)
- fix(config): the Runtime line at the top of first-run setup promises only the PATH fix the end shows (#2510)
- fix(protocol): a step reads only the stage protocol parts its branch can take (#1330) (#2513)
- fix(review): a review that did not finish says so at the gate, and why only when the conductor knows (#2508)
- fix(build-and-test): the CI target is CI Pipeline's while CI Pipeline is still ahead (#2509)
- fix(engine): a setting reads "set by you" only when a message of yours asked for it (#2468)
- fix(engine): a stage you skip where it stands names its own command to run it again (#2507)
- fix(review): a review reads only the review branches it can take (#1330) (#2499)
- fix(session): on a busy Mac, /aidlc intent and SessionStart find their chat again (#2500)
- fix(engine): a skip you ask for while Construction runs one Unit at a time is carried out, for that Unit or for every Unit (#2485)
- fix(construction): Functional Design reads the stories whose acceptance criteria it traces (#1086) (#2486)
Source commit: 9ecefa0
Update from the last release
Updating from the last release to this preview failed in the preview run.
Nightly test report
Full Suite failed for 9ecefa0dda7f in run 38809298225. The preview build does not wait for these tests.
Failed legs: live_windows (failure).
Failing tests
t-tui-workshop.serial(1 failed assertions) inlive-claude-tui-2-Windowst-tui-workshop (answering AUQ gates advances disk state) > workshop run-through commits affirmation on disk
Failed jobs
- Linux / result (1): Linux / result
- Windows / claude-tui (1): Windows / claude-tui 2/3
Full logs are in the run 38809298225 full-suite-* artifacts.