Public: anyone can read this; nobody can change it.
A crew clone's own exclude file keeps what is never the work out of every commit #22
Check · Review · Land done · Live — Landed; nothing deploys from this branch, so this is the end of its path.
Two crew pulls on the fork today (trevato/forge#1130, #1132) carried a `.tmp/` directory: bun's install cache and a test forge's data directory, 584,000 lines, committed by the wrap-up as "leftover changes" through `git add -A`. Inside the shell sandbox the temp directory sits beside the code, so anything a run installs or tests lands in the tree, and the wrap-up faithfully commits it.
shipped merged
Sign in to join this conversation.
new above
offline
Two crew pulls on the fork today (trevato/forge#1130, #1132) carried a
.tmp/directory: bun's install cache and a test forge's data directory, 584,000 lines, committed by the wrap-up as "leftover changes" throughgit add -A. Inside the shell sandbox the temp directory sits beside the code, so anything a run installs or tests lands in the tree, and the wrap-up faithfully commits it.The clone now carries its own
.git/info/exclude, written by the forge inprepareClone:.tmp/andnode_modules/are never the work, in any repository the crew clones, without touching the repository's.gitignore.test/crew.test.ts's stand-in agent now plants.tmp/cache/leftoverbefore itsgit add -A, and the test proves the pull carriesCREW.mdand nothing under.tmp/.Why the temp directory sits inside the clone at all (Claude Code's sandbox and
CLAUDE_CODE_TMPDIR) is for trevato/forge#1107; this keeps the damage out of git either way.Typecheck clean;
crew,crew-wrapup,hollowgreen locally.🤖 Generated with Claude Code