seed-projectslisted

# seed-projects After the user has approved an import plan (the output of `infer-vault-structure`), someone has to actually write the files. Writing them naively means dozens of preview prompts (one per file), which causes user fatigue and rubber-stamping. Writing them without a preview is unsafe — the user may want to edit a routing decision or abort entirely. This skill solves both: stage every file under a tmp directory, render ONE unified diff bundle covering the whole batch, prompt the user once with `[a/e/s/b]`, and on apply walk the staging tree with atomic `cp + mv` per file. If apply fails partway through, already-written files round-trip through the preview as already-existing targets on re-run, so resuming is safe. Each generated file carries provenance frontmatter so downstream tools can detect auto-generated artifacts and avoid double-writing. ## Invocation `/seed-projects --vault-root <path>` — the `/adopt` skill auto-invokes this when an approved import plan is present at `$CLAUDE_HOME/onboarding/seed-content/state/approved-import-plan.md` (post-install) or the foundation-repo equivalent (during testing). Direct invocation: ```sh ./seed.sh \ --vault-root /Users/me/MyVault \ --approved-plan onboarding/seed-content/state/approved-import-plan.md # → /Users/me/MyVault/Engagements/<label>/{PRD.md, Context.md, Updates.md} # → /Users/me/MyVault/Inbox/<date>-<slug>.md per non-project candidate # → audit log entries appended at onboarding/auto-author-log.jsonl