Description
When SessionSpec.expected_spec is not set, _capture_launch_auto_run_results (and the harvest that follows) finds result specs with a non-recursive glob:
# adapters/generic.py:1410 (0.11.1, unchanged on main)
for artifacts in self._artifact_dirs(spec.cwd):
paths.extend(artifacts.glob("*.md"))A project that keeps story specs one level down (e.g. _bmad-output/implementation-artifacts/stories/<key>.md) gets no-artifact on every Stop, with nothing in the message saying which directory was searched or that subfolders are not.
This is a different path from #224 (spec present but missing its ## Auto Run Result marker). Here the spec is complete; it is just not in the directory the glob looks at.
Steps to reproduce
- Put a story spec under
implementation-artifacts/stories/ instead of directly in implementation-artifacts/.
- Run a story where
expected_spec is not set.
- Every Stop reports
no-artifact; the session rides to timeout.
Moving the spec directly under implementation-artifacts/ makes the same run succeed.
Suggested fix
Either of these would have saved a long debugging session:
- search recursively (
rglob("*.md")), or
- keep the non-recursive glob but name the searched directory in the
no-artifact breadcrumb, e.g. no-artifact: no *.md directly under <dir> (subfolders are not searched).
The second keeps current behaviour and only makes the failure self-explaining, if the flat layout is intentional.
Environment
Windows 11, bmad-loop 0.11.1 (uv tool), claude adapter, BMAD v6.12.0.
Description
When
SessionSpec.expected_specis not set,_capture_launch_auto_run_results(and the harvest that follows) finds result specs with a non-recursive glob:A project that keeps story specs one level down (e.g.
_bmad-output/implementation-artifacts/stories/<key>.md) getsno-artifacton every Stop, with nothing in the message saying which directory was searched or that subfolders are not.This is a different path from #224 (spec present but missing its
## Auto Run Resultmarker). Here the spec is complete; it is just not in the directory the glob looks at.Steps to reproduce
implementation-artifacts/stories/instead of directly inimplementation-artifacts/.expected_specis not set.no-artifact; the session rides to timeout.Moving the spec directly under
implementation-artifacts/makes the same run succeed.Suggested fix
Either of these would have saved a long debugging session:
rglob("*.md")), orno-artifactbreadcrumb, e.g.no-artifact: no *.md directly under <dir> (subfolders are not searched).The second keeps current behaviour and only makes the failure self-explaining, if the flat layout is intentional.
Environment
Windows 11, bmad-loop 0.11.1 (uv tool),
claudeadapter, BMAD v6.12.0.