upstream bmad-build-auto rename (bmad-method >= 6.10.1-next.33) breaks validate/run/sweep

[pbean]

Upstream BMAD-METHOD PR #2651 (first shipped in bmad-method@6.10.1-next.33, 2026-07-30)
renamed the dev-primitive skill bmad-dev-auto → bmad-build-auto, leaving a forwarding
shim named bmad-dev-auto (a lone SKILL.md with an interactive customization-migration
gate that can HALT unattended sessions with zero disk writes). On an upgraded target
project, bmad-loop validate fails skills.base-incomplete (the shim lacks the
step-04-review.md + customize.toml markers), the same preflight hard-blocks
run/sweep/resume, and worktree isolation would never copy bmad-build-auto.

The rename itself is contract-clean (spec status: flip, ## Auto Run Result,
"folder+id dispatch" marker, terminal statuses, Stop-event completion all unchanged; only
the no-spec fallback result filename changed to bmad-build-auto-result-*). Two other
upstream changes rode the same window and also break bmad-loop:

• PR #2601: SKILL.md is now a renderer stub running _bmad/scripts/render_skill.py
  via uv — HALTs with no spec write when the script is missing. Step files and
  customize.toml still exist in the skill dir (markers remain valid).
• PR #2640: deferred findings now land in the spec's frontmatter deferred: YAML
  list, not deferred-work.md — the sweep pipeline silently starves.

Everything on main since the v0.9.0 tag is next-major work and must not ship, so the fix
is cut as hotfix 0.9.1 from tag v0.9.0 (commit 7255174) on a release/0.9.x
branch. Upstream npm latest (6.10.0) is still pre-rename, so 0.9.1 must keep working
against both skill eras.

Scope of the fix

1. Resolve the primitive on disk — prefer bmad-build-auto, fall back to a
   marker-complete bmad-dev-auto; never invoke or accept the shim. New validate checks
   skills.base-shim, skills.dev-renderer, skills.customize-legacy.
2. Thread the resolved name through every session prompt (dev, review, repair,
   stories dispatch, sweep bundles) and the no-spec fallback result marker.
3. Harvest spec-frontmatter deferred: findings into the deferred-work ledger,
   era-agnostically (gated on content presence, never on the installed skill name).

Delivery: one hotfix PR into release/0.9.x, then a forward-port PR to main.

Out of scope (follow-ups)

• uv-on-PATH probing for the renderer stub (only script presence is probed).
• Seeding gitignored _bmad/custom/* into isolation worktrees (profile seed_files
  covers this case-by-case).
• Sweep consuming spec-frontmatter deferrals directly / upstream structured-sink work
  (relates to deferredwork: dev session writes ledger markdown directly — give it a structured sink the orchestrator renders #326).
• If the 0.9.x line continues after the next major ships: release.py last_release_tag()
  is version-ordered, so a 0.9.2 prepared after 0.10.0 would diff TUI assets against the
  wrong base.

[coderabbitai]

⚠️ Possible Duplicate Issue(s)

• [BUG] validate and run blocked: bmm >=6.10.1 renamed the dev primitive to bmad-build-auto, leaving bmad-dev-auto a shim #393

🔗 Related PRs

#56 - chore(release): 0.8.0 [merged]
#73 - fix(setup): never delete live core BMAD config/manifest (#64) [merged]
#86 - feat(resolve): intent-gap patch-restore re-drive + inline-review-layer preflight (re-cut) [merged]
#247 - chore(release): 0.9.0 [merged]
#283 - fix(validate): accept merged bmad-review, drop unreleased verification-gap requirement (#260) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!