Repository navigation
verify-package-attribution.sh: respect CARGO_TARGET_DIR - #21
Conversation
A gate that gives each worktree its own target directory packaged the facade successfully and then failed the attribution gate, because the script looked only in ./target. Found running ci-local.sh on eigen. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Main passed every gate at cf7e2a4. At posting time main is ff9524d and the only code that moved is ci-local.sh, which is #19 merging and carries its own verdict from this host, so main code is verified as of this line. Notes that scripts/ will match on every CI merge, so the recheck wants reading rather than a clean-or-dirty answer. G5 started on PR #21. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Host grust (Linux x86_64), full gates, own detached worktree at the branch head. Code drift check at posting time: main is What this verdict does not cover. The bug #21 fixes appears only when This is G5 from task board #5. |
PR #21 passed every gate at d499128. The verdict states what it does not cover: the bug needs CARGO_TARGET_DIR set and this host isolates by worktree instead. G5 package gate again resolved every sibling from tmp-registry with no crates.io lookups, matching G0, which makes the host difference environmental and supports the cached-tarball explanation for quegee failure. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
1 participant
scripts/ci-local.shrun withCARGO_TARGET_DIRset fails its last gate:cargo packagehad succeeded and written the archive to the configured target directory; only this script's hardcodedtarget/package/...was wrong. Found on eigen, where each gate worktree gets its own target directory so that concurrent gates do not share one.One line:
target=${CARGO_TARGET_DIR:-target}. Unchanged for anyone who does not set the variable, which is why neither Linux host hit it.🤖 Generated with Claude Code