When learn runs against a vault with no progress marker (e.g. a fresh/empty
vault), engram transcript --mark exits non-zero and the learn SKILL.md
instructs stopping to ask the user via AskUserQuestion which date to scan from.
In any non-interactive / headless context (claude -p, automation, the eval
harness's cold arm) there is no user to answer — the step hangs, errors, or
silently skips capture. This blocks any automated learn-on-empty-vault flow.
Discovered while designing the self-seeding eval harness (see
docs/superpowers/specs/2026-05-29-memory-eval-harness-design.md, Run-1
blockers). Fix likely: a non-interactive default / flag for the first-run scan
(e.g. --from all or a marker bootstrap) so headless learn can proceed.
When
learnruns against a vault with no progress marker (e.g. a fresh/emptyvault),
engram transcript --markexits non-zero and the learn SKILL.mdinstructs stopping to ask the user via
AskUserQuestionwhich date to scan from.In any non-interactive / headless context (
claude -p, automation, the evalharness's cold arm) there is no user to answer — the step hangs, errors, or
silently skips capture. This blocks any automated learn-on-empty-vault flow.
Discovered while designing the self-seeding eval harness (see
docs/superpowers/specs/2026-05-29-memory-eval-harness-design.md, Run-1blockers). Fix likely: a non-interactive default / flag for the first-run scan
(e.g.
--from allor a marker bootstrap) so headless learn can proceed.