adopt-project.py execute exits 2 with nothing on stderr when git filter-repo fails, after both leg remotes exist: _build_the_legs swallows CommandFailed on the premise that run already reported it, and run only raises #171

(lane openRepoShape-2, 2026-10-09T02:08:54Z) Found by the re-verification of the #166 branch at 09d7889; pre-existing on main 7f84ca4 and outside #166's scope. Not claimed.

What is wrong. _build_the_legs catches CommandFailed from the leg extraction and returns None (adopt-project.py:1652-1653 at 09d7889; the same except CommandFailed: return None on main). Its docstring says run has already said which command failed and why, but run (scripts/shape_materialize.py:474-481) only raises. cmd_execute then returns 2. The human sees the two bare .../Northwind-spec.git / Northwind-code.git lines on stdout, exit 2, and an EMPTY stderr. Both leg remotes exist afterwards, so a re-run meets leg-remote-exists.

Reproduction (any source whose .gitmodules git's --list accepts but whose checkout git's submodule reader dies on; on main an unterminated section does it, and on 09d7889 a bare path key in a second section):

printf '[submodule "a"]\n\tpath = upstream/dependency\n\turl = %s\n[submodule "b"]\n\tpath\n' "$DEP" > Thing/.gitmodules   # plus the gitlink at upstream/dependency
adopt-project.py plan --source Thing --project Northwind ... ; resolve .gitmodules and upstream/ to code
adopt-project.py check --plan plan.yaml                         # exit 0 at 09d7889
adopt-project.py execute --plan plan.yaml --yes --local-remote-dir remotes --work-dir work > out 2> err; echo $?   # 2
wc -c err                                                       # 0
ls remotes                                                      # Northwind-code.git  Northwind-spec.git

With CommandFailed instrumented, the swallowed failure is git filter-repo --paths-from-file .../spec-paths.txt --force with stderr error: missing value for 'submodule.b.path' / fatal: bad config line 5 in file .../Northwind-spec/.gitmodules / Error: fast-export failed. A symlinked .gitmodules with everything dropped reaches the same silent exit on both trees.

What would fix it. Print the failure (exc.loudly(...), or whatever run's callers use elsewhere) before returning None, naming the command and its stderr, so an execute that stops after creating the legs says why. Independent of #166, #165, #167 and #168; a one-hunk change in _build_the_legs plus a test that asserts stderr names git filter-repo when it fails.

Driver and probe: opensoft/brett-wip handoffs/xFactory/attachments/lane-openRepoShape-2/review166-09d7889/ (r3-report.md N4, r3tools/runexec2.py).