Warning: Full Suite failed for this source. A Full Suite failure report ends these notes.
Merged commits since preview source f7af394:
- fix(compose): an approved tailored plan is created with every stage the person approved (#2923)
- fix(gates): a stage gate records the choice the agent read from the person's reply, whatever they typed, with their words kept whole (#2907)
- feat(reverse-engineering): ask once before reading ten or more repositories, and take the list in one command (#2919)
- fix(codex): an empty structured-question box falls back to numbered prose automatically (#2782)
- fix(gates): the gate reply says when Accept as-is is on offer, so a park keeps it (#2905)
- fix(guards): the Code Generation gate comes back after a reviewed file changes at it, instead of a refusal loop (#2916)
- feat(reverse-engineering): say how many repositories the analysis will read (#2881)
- fix(reverse-engineering): a workspace of many repositories starts Reverse Engineering and reaches its gate (#2918)
- test(live): live checks judge Windows runs right and meet the blocked-sensor gate as a run leaves it (#2917)
- fix(construction): a Redo naming the Unit in flight redoes its step instead of routing back to it (#2772) (#2886)
- fix(intent-capture): the agent tags the workflow scope only in Initial Scope Signal, so the gate no longer asks to fix its own tags (#2915)
- fix(review): the review summary says what the review found, in one voice (#2914)
- fix(config): a check another settings file still keeps off is said to be still off, by its own name (#2913)
- fix(ci): release and Full Suite steps run on Node 24, so the nightly preview no longer warns about Node.js 20 (#2912)
- fix(state): running one stage on its own with --single leaves your work where it was (#2911)
- fix(stop-hook): in a run you are driving, the turn ends where the agent waits for you, with no carrying-on line (#2903)
- fix(construction): a Unit whose step is longer than one command result starts, instead of being refused every round (#2908)
- fix(traceability): requirement IDs are traced as Requirements Analysis wrote them, later tables start from the tool, and the check reads the finished files at the gate (#2906)
- fix(brief): a delegated agent does not move the person's working tree or install unnamed packages (#2812)
- docs: the contributor guide and both rule-loading sentences name every harness, Cursor included (#2794)
- test(drivers): the test person picks an option by what it says, and the live checks read the record, not the words (#2902)
- fix(presence): a prompt Kiro or opencode made is told from the person's by the host's own mark, not by its words (#2901)
- fix(scope): a switch or a jump past a Units Generation that never ran goes on with the stages it has (#2899)
- fix(copilot): the person's redo or change for every Unit, and one Unit's change, runs the reopen next --stage prints, instead of being refused every round (#2888)
- fix(gates): Accept as-is comes on offer after three change requests on the same stage, not across the workflow (#1083) (#2884)
- fix(review): a finding a review raises after a Unit's approval is recorded as Approved autonomously, not as a risk the person accepted (#2900)
- fix(guards): name the Request Changes report (#2700)
- fix(sensors): nothing sends the agent to run a check AI-DLC runs itself, and engine sensors and a check script's --help name the command (#2882)
- fix(plugin): removing a plugin or a config refresh keeps a stage line the person wrote beside the plugin's own (#2896)
- fix(team): a team Unit whose stage is done is named the step it still owes, not its stage again (#2891)
- test(harness): every live leg keeps its projects' audit and what the agent told the person, and the trace audit reads them (#2885)
- fix(guards): every refusal path writes its audit row, with a reason (#2835)
Source commit: 3b3beb5
Nightly test report
Full Suite failed for 3b3beb58b49d in run 39856584497. The preview build does not wait for these tests.
Failed legs: live_windows (failure).
Failing tests
No failing test files were recorded in the downloaded evidence. The failed jobs below may have stopped before or outside their tests.
Failed jobs
- Linux / result (1): Linux / result
- Windows / claude-tui (1): Windows / claude-tui 1/3
Full logs are in the run 39856584497 full-suite-* artifacts.