[beta.2.3] Simplify CI and AutoFlow around final semantic responsibilities

[SisyphusZheng]

ADR-0152 / #1341 scheduling amendment: execute cleanup alongside every Beta.2.1/Beta.2.2 replacement; Beta.2.3 is the residual closure checkpoint, not the start of cleanup. Remove displaced owners and their callers/checkers/docs together with regression evidence.

[beta.2.3] Simplify CI and AutoFlow around final semantic responsibilities

Part of #1322. Release: beta.2.3.

Principle

Optimize for responsibilities, not workflow file count.

Conceptual CI areas

PR qualification; release; security; published consumer qualification; scheduled stress/reporting where truly appropriate.

AutoFlow owns (OE-specific invariants)

Compiler/runtime parity; Part Program; SSR/claim/islands; Router semantics; Document semantics; zero-JS admission; package/public surface; artifact qualification; release-state transition.

AutoFlow does NOT own (generic — mature tools do)

Formatting; Markdown grammar; Action syntax; secret scanning; dependency updates; workflow count; generic package metadata lint.

Specific items

• Delete duplicate gate wrappers where plain CI tool execution already provides the authoritative check.
• Re-adjudicate check-workflow-slimming (workflow-count policy: delete), check-action-pins (keep only OE-specific release-credential/provenance assertions if Renovate/zizmor cover the rest), check-role-neutral-docs (retire if it only serves the temporary execution scheme), check-repo-hygiene (decompose by responsibility — no miscellaneous policy bin), check-architecture-contract (keep real invariants; remove expired migration exceptions, generic style policy, historical absence assertions, obsolete revisitBy allowances).
• Consider merging the two manual smoke workflows (supabase-project-smoke + fullstack-deploy-smoke, 988 lines, overlapping responsibilities) into one parameterized workflow.
• Do not add a new governance framework.

Release

beta.2.3.

Superseded — Accepted refinement 2026-09-08 (partially superseded 2026-09-09)

The 2026-09-08 version made a final-HEAD AI review an IMMEDIATE MERGE PREREQUISITE for
#1343. That merge-blocking requirement is superseded by the 2026-09-09 maintainer
decision below. What was implemented from it and remains: provider/API failure
detection and visible reporting (tools/check-review-outcome.ts + opencode-review.yml,
with unit tests for the error and valid-review paths) — a green wrapper is never
counted as review evidence. What was superseded: failing required CI / blocking merge
solely because the supplemental review provider is unavailable.

Governing plan (2026-09-08)

Maintainer-approved: ADR-0152, docs/current/VERSION_PLAN.md and #1341. Execution: https://github.com/orgs/open-element/projects/3. Acceptance checkpoints replace the three-day schedule. No automatic beta.2.4. Native/Lit are formal Framework Mode targets, not already-shipped support. Existing release records/versions remain unchanged; no closure without final-SHA evidence. examples/** stays frozen.

Accepted upstream-contribution priority — 2026-09-08

Prefer standard runner and direct CI execution; prepare upstream-native reproductions for generic WTR/Playwright failures. Remove redundant critical-path spawning/output interpretation only after nonzero failure behavior is preserved. The #1343 insufficient-balance review finding was resolved by detection/reporting; merge-blocking was superseded 2026-09-09 (see below).

Use docs/architecture/infrastructure-reduction.md and VERSION_PLAN as scope. Prepare useful generic fixes for contribution through normal authorization; no invented bugs, cosmetic PR quota, or upstream-response release dependency. No migration is completed by this update.

Current policy — maintainer decision 2026-09-09 (authoritative)

• Third-party AI review is supplemental and non-blocking.
• Provider/API errors are still detected and surfaced (warning annotation on the job,
  never a silent green, never counted as review evidence).
• A provider failure (e.g. the observed APIError: Insufficient Balance) contributes
  zero independent review evidence.
• Required CI, the fresh verifier and human review remain the actual merge gates.
• Restoring provider availability remains a separate external/billing concern; do not
  switch provider/model or spend money without authorization.

Approved executable maturation — 2026-09-09

• Required WTR jobs must fail on missing browsers, setup/spawn errors and assertion failures. Reuse official browser launcher and compiled artifacts.
• Remove displaced runner/checker wrappers only after equivalent failure propagation.
• Preserve the feat: implement Beta.2.1 unified Router and standalone Element tooling #1343 issuecomment-5579272257 obligation: verify visible provider-error reporting and required CI/fresh verifier/human review; wrapper success is not review evidence. Recheck live status rather than treating the historical failure as permanently current.

Scope: docs/architecture/alpha-maturation.md and VERSION_PLAN (planning PR #1342). Upstream contribution potential can justify bounded extra integration cost; no invented defects, PR quota or upstream-response release dependency. This update changes planned work, not implementation status.

Current supplemental review policy — confirmed 2026-09-09

Supersedes earlier AI-review availability merge prerequisites in this issue. Third-party AI review is supplemental and non-blocking. Report provider/API errors visibly; wrapper SUCCESS is not completed review and provider failure contributes zero independent review evidence. Required CI, fresh verifier and human review remain the merge gates. Retain #1343 (comment) as historical evidence. No provider/model switch or billing action is authorized.

[SisyphusZheng]

Update from the Beta.2.1 implementation session (PR #1343, HEAD bf7d00b):

1. Detection/failure propagation is fixed: tools/check-review-outcome.ts + the opencode-review workflow now verify the review step outcome and the run-attributed PR comment against provider/API error signatures; a green wrapper can no longer masquerade as completed review evidence. Unit tests cover the genuine-review and each error path (12 tests).
2. Maintainer decision 2026-09-09: the review job is NOT a CI blocker. The gate's verdict is therefore reported as a warning annotation, not a red check. Verified live on bf7d00b: the provider still posts APIError: Insufficient Balance, the gate reports it, job concludes success.
3. Drift to reconcile on the planning side (not edited by the implementation branch): VERSION_PLAN.md and ADR-0152 still say the review failure must be resolved before merging feat: implement Beta.2.1 unified Router and standalone Element tooling #1343 / fail closed. Those belong to open planning PR docs: plan Beta.2 preparation and upstream-oriented 1.0 Alpha #1342; syncing that wording to the non-blocking decision is the planner's edit, not mine.
4. Restoring actual review capacity (DeepSeek balance) remains an external billing action. This issue stays open for its Beta.2.3 CI-simplification scope.