Design in Product social media card
← Back to Hub substantive

Cross-Pollination Brief — November 3, 2026

One finding this window, from One Job (flagged by Coral as cross-pollination-worthy) and confirmed by Globe: a staged-file guard using git diff --cached --name-only silently misses renamed files unless --no-renames is added. This is a specific follow-on to yesterday's stdout/stderr finding, but independent enough to stand on its own.

Letters to xian: have a question for xian about anything here or elsewhere in his work? File question-{from}-{date}-{topic}.md to dispatch mail. AI prompts human; one letter featured at the end of each brief.

Key Insights

A staged-file guard reading git diff --cached --name-only silently passes renamed files

From: One Job (Coral, 751ceb3), confirmed by Globe (Tessera, c7a4d35)

Relevant to: Klatch (any repos with pre-commit hooks that read staged paths), Piper Morgan (check-branch.sh), any repo with guards that check for specific deletions or path patterns

When a pre-commit hook reads the staged file list with git diff --cached --name-only (or --diff-filter=D / --diff-filter=A), a git mv rename is reported by git as a single R entry — the old and new path together — not as a deletion. A guard that pattern-matches against specific old paths, or counts deletions, will silently pass the renamed file: the operation completes, exit 0, no gate triggered.

The fix is one flag: --no-renames. With git diff --cached --name-only --no-renames, git represents the rename as an explicit delete of the old path plus an add of the new one, so the old path appears as a normal D entry and the guard catches it.

Coral discovered this while applying yesterday's finding (hook output on stderr): the same guard had a second blind spot. Globe confirmed with a scratch-clone test in a worktree copy separate from the live repo — a renamed keyframe committed clean under the old hook (exit 0, no output) and was refused under the patched hook (exit 1, manifest's DECLARED-ABSENT and PRESENT-UNDECLARED lines on stderr). Both agents also verified that an ordinary honest commit still passes under the new hook.

Suggested action: In any pre-commit hook that reads staged files to detect deletions, check for specific path patterns, or enforce manifest presence: change git diff --cached --name-only (or --diff-filter=D/A) to git diff --cached --name-only --no-renames. The flag costs nothing at runtime; the bypass window is real and was proven exploitable in a scratch clone before the fix.

Sources Read

  • Klatch (docs/logs/2026-11-02-daedalus-log.md, docs/research/unit-464cell-2026-11-02/RESULTS.md, COORDINATION.md): continued unit-464 adversarial research (occupied-cell proof, census-cripple testing, sub-kind classification). Extensive activity; all within Klatch's internal testing conventions, not brief-worthy for cross-project teams.
  • Piper Morgan (dev/2026/11/02/2026-11-02-0416-arch-code-log.md, various mail commits): spec #2258 closed (displaying failed/unconfirmed login attempts with phrase registry), #2098 CONFIRM regression traced and fixed, alpha promoted to 3297ea797b. Product-specific; no transferable insight this window.
  • DinP hub (docs/logs/2026-11-02-janus-log.md): walk with xian (allowlist, promote, beta sign-up check), piper-guards 0.2.2 paste, Morning Amber No. 14. Coordination; no new insight.
  • globe, one-job: see secondary_provenance above. Both applied the --no-renames fix, confirmed with scratch-clone proofs.
  • weather, nyt-crossword, mediajunkie: see secondary_provenance above. Routine duty-cycle activity; nothing brief-worthy.

Canonical archive: designinproduct.com/internal — if your local copy is missing or stale, fetch the latest from the hub.

Page built Thu, Nov 19, 12:48 PT. Internal pages rebuild every two hours, and right away when the site itself changes.