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
Resuming a crew-afk sprint from wherever the shell happened to be sitting could silently
adopt the wrong branch as the feature branch: session-init.sh's default-branch heuristic
only checked whether the current branch equaled the repo's default branch, so re-running with a
slug that already had a .scratch/<slug>/sprint.env from an earlier session — while sitting on
an issue branch left over from a prior round, a detached HEAD, or a colleague's branch — read
"not on default" as "already on the right one" and never switched to the recorded FEATURE_BRANCH. Fixed by pinning to the session's recorded feature branch first, checking it
out if the shell isn't already there, before either branch-setup path runs.
A reused issue branch (this round's resume, or a worktree left over from an earlier crew-afk invocation) could fork from the feature branch before other issues merged into it,
leaving the coder to work from a stale base and the gap to surface as an unexplained conflict at
the merge gate ~45 minutes later. ensureWorktree now flags a reused branch, and runWorker
forward-merges the feature branch into it inside its own worktree before the coder starts; a
merge conflict aborts cleanly and reports the issue as blocked instead of masking it.
gen-override.sh's per-ecosystem passthrough vars were named and commented as if they only
ever carried proxy settings, which obscured that the same bare-name convention also covers
non-proxy secrets — Node's NPM_TOKEN for private-registry auth during npm install was
missing from the list entirely. Renamed ECO_PROXY_VARS to ECO_ENV_PASSTHROUGH and added NPM_TOKEN to the Node ecosystem's list.