You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Added
A workflow ## Environment heading: how to set up and run apps and tests,
where sign-in data comes from, how to check and refresh credentials, and
when to stop. Task and dev sessions read it whenever they run something.
An environment log in integration.md, and a "Done, try again" question for
the few things only you can do (a browser sign-in, sudo, a closed test
environment).
Word targets for the brief, the spec and the plan.
Hook reminders: Integration that has not run yet, and questions still
waiting for an answer; the language reminder also follows subagent reports.
A status line an answer writes (Approved:, Cleanup:, Push:, CI:)
counts as that answer.
The push question shows each repository's branch and base in a table.
Checks that wait for a later step: name them under ## Push and CI is
followed without them, so ## Push:After (for example a QA hand-off) can
run. ci_wait.py watches the other checks.
Dev sessions ask for blockers with the same "Done, try again" question.
Changed
Every closed question has buttons, the recommended answer first, after a
short box saying what is needed, why, and what each answer does. A question
interrupted by another request is asked again.
Approvals stay one at a time, and reviews are lighter: one round, a second
only for fixed critical or major findings, at most three per document. A
later change is sorted first; only a critical change gets a review and a new
approval.
Worktree setup, a setting changed for one test, and new test data named with
the task key no longer ask. After Integration one question asks whether to
keep or remove the test data.
Integration always runs after the last repository is done, and there is no
push question before it.
A change you ask for after an approval counts as approved; Rabbit asks again
only for a critical change it makes itself.
The setup questions use buttons too.
The observer writes its findings in the document language.
impact.md and the spec list only surfaces that cross repositories; with
none they say so in one line.
Shorter Scenarios: cases with the same result share one row.
Removed
The continue and yes native answers, and the question before installing
a worktree's dependencies.