PR #155 disposition: partly superseded by the 0.9.1 renderer port, needs reconciling on main

[pbean]

Context

Follow-up from #405 / PR #406.

PR #155 (fix/renderer-era-dev-auto, draft, open since 2026-07-15, last touched 2026-07-23)
targets main. PR #406's review round ported the load-bearing half of it onto the 0.9.1 hotfix
branch — adapted, not cherry-picked — and deliberately cut the rest. #155 is therefore now
partly superseded and partly wrong, and it must be reconciled during the release's forward-port
to main rather than merged as-is or closed silently.

What landed in 0.9.1 (adapted from #155)

• _seed_bmad_tree — whole-directory per-file merge of the repo's _bmad/ into the
  worktree, copy-when-absent, BMAD_SEED_EXCLUDES = ("render",) skipped before descending, the
  seed_files loop's resolve-and-contain symlink guard.
• /_bmad/render/ exclude inside the worktree, and _bmad/render/ in init's gitignore tuple.
• _bmad_scripts_seed_incomplete — new, fix(worktree): survive the renderer-era bmad-dev-auto (BMAD-METHOD #2587/#2588) #155 could not have needed it; reports a short
  _bmad/scripts/ seed through the existing worktree-seed-skipped channel, as the exported
  BMAD_SCRIPTS_SEED_REL sentinel. The engine escalates on that one entry and pauses the run
  (added in the same review round that reversed the _require_base_skills cut, below): the
  fault is identical for every story, so dispatching would burn the backlog on result-less
  Stops.
• _require_base_skills now hard-blocks the two renderer checks — see the reversal note
  under the cut table.

What was cut, and why it must not be revived unchanged

Cut	Reason	Tracked as
--render-probe / _run_render_probe	Guesses an argv the agent actually substitutes; green would be a lie	#407
toml_artifact_parity, _TOML_LAYERS, _toml_merge	Model is wrong now: ambiguity is a RenderError, not "bmm wins"; plain deep-merge vs structural_merge	#408
tracked_render_warning	Hygiene, not a HALT	#409
renderer_shim_trees	Hardcodes the legacy skill dir; this branch resolves per tree via resolve_dev_primitive, and keying on the name violates the era-agnostic rule	—
missing_renderer_support's uv-on-PATH FAIL	A validate-host PATH probe does not answer the session-PATH question (fresh panes); the stub already reports uv-missing loudly	—
missing_renderer_support's "render.py next to SKILL.md" FAIL	Same condition as the shipped skills.dev-renderer, relocated because upstream moved the script — porting it would double-report	—
One-line addition to _require_base_skills	REVERSED — now taken, see below	—
CONTRIBUTING.md paragraph	Asserts uv is checked, which we deliberately do not	—

Reversal (2026-07-30): the _require_base_skills row is now TAKEN

Codex re-reviewed PR #406 and refiled this, and the cut rationale did not survive inspection.
"Newly aborting projects that pass today" means projects that pass validate, not projects
that work: a resolved renderer stub with no _bmad/scripts/render_skill.py (or no
_bmad/config.toml) HALTs every unattended session, so the green preflight was promising a
run that could not possibly produce a result. And both check ids are new in 0.9.1 and
unreleased, so promoting their severity before the release costs no shipped behaviour —
there is no project in the field whose passing run this takes away.

The shape differs from #155's one-liner, so do not port that either: _require_base_skills has
no severity filter at all — it FAILs on whatever missing_base_skills returns — so the two
emitters moved out of dev_primitive_warnings into missing_base_skills at severity
problem. dev_primitive_warnings keeps only skills.customize-legacy, which is genuinely
survivable (the session runs, unstyled). run --dry-run's "NOT runnable as-is" banner reads
the same list and picked both conditions up for free.

RENDERER_PROBE_TEXT = "render.py" in #155 is also dead on arrival: it is not a substring of
render_skill.py, so its era gate would silently never fire. install.RENDERER_SCRIPT_MARKER
is the correct marker.

Forward-port hazards (main has moved under #155 and under 0.9.1)

1. Provisioning was rehomed. provision_worktree now lives in worktree_flow.py:70 on
   main (Engine collaborator extraction: worktree + rollback clusters, rehome provision_worktree #244 collaborator extraction), not install.py. The 0.9.1 patch is against
   install.py and will not apply.
2. The exclude mechanism changed shape. _worktree_local_exclude on main returns
   str | None and, since provision_worktree's exclude patterns are repo-wide and permanent: new files under .claude/skills silently stop being staged in the main checkout #384, writes to the worktree-scoped
   .git/worktrees/<id>/info/exclude activated by a worktree-scoped core.excludesFile — not
   the common .git/info/exclude. The 0.9.1 gate on /_bmad/render/ ("only when the worktree
   has a _bmad/, because the exclude file is shared with the main checkout") is reasoned from
   the old mechanism. On main the pollution argument no longer holds, so re-derive the gate
   there instead of transplanting the condition — and keep its ablation
   (…render_shield_only_when_bmad_present) honest to whichever rationale survives.

Disposition

Decide one of: rebase #155 onto post-forward-port main reduced to only what was not ported,
or close it with a pointer to #407/#408/#409. Either way it should not stay open as an unrebased draft
that reads as "the renderer fix, still pending".

[coderabbitai]

🔗 Related PRs

#95 - feat(adapters): post-kill reconcile rescues finished stalled/timeout sessions [merged]
#113 - fix: degrade spec read faults instead of crashing the run [merged]
#155 - fix(worktree): survive the renderer-era bmad-dev-auto (BMAD-METHOD #2587/#2588) [open]
#277 - feat(adapters,engine,devcontract): deterministic missing-marker catch + repair [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!