governance: eliminate repeated AI stop and confirmation loops

[wpessential]

Root causes fixed

Fresh audit found four repository-native causes of repeated stop/question behavior:

1. AGENTS.md still treated closed historical Issue NEXT-AUTH — Define post-M13 WorkIntel product authority #50 as the current M14 product-authority gate.
2. Start/resume protocol literally said stop and rehydrate when main/head/scope/governance moved.
3. .ai/NEXT-ACTION-OPTIONS.md forced URL-only repository entry into a read-only 1–3 option prompt.
4. Committed README/compact state carried volatile PR/branch/merge transaction facts, causing post-merge semantic drift and recursive reconciliation PRs.

Changes

• closed/superseded authority records cannot trigger owner confirmation;
• material Git/governance drift is rehydrated inside the same execution window;
• routine technical work, task selection and fallback selection remain AI-owned;
• URL-only repository entry resumes safe authorized work automatically;
• numbered engineering choices are disabled by default and shown only when the user explicitly asks for options;
• user input is requested only after all safe lanes are exhausted and a genuinely user-owned fact/new scope/legal-commercial/protected authority is required;
• committed compact state/README are semantic checkpoints, not live PR transaction ledgers;
• README AI progress excludes Active PR, Active Branch and Last Completed transaction fields;
• live PR/branch/check/merge facts come from GitHub at runtime;
• post-merge state-only reconciliation is explicitly not required;
• deterministic claims, tests and supervisor audit ratchets enforce these rules.

Safety

Exact-head CI/review, security, secrets, product-scope, destructive actions, migrations, provider/production/release authority and Runner authorization remain fail-closed. This removes conversational fragmentation; it does not broaden protected authority.

Verification

Fresh exact-head WorkIntel CI, Code Quality and Windows Certification are required before merge.

[wpessential]

Fresh repeated-stop audit completed on exact head 24e2590fe7428822d3aa554cdb4a93ef9da6d1c9.

Root causes addressed:

• stale closed Issue NEXT-AUTH — Define post-M13 WorkIntel product authority #50 no longer acts as a live approval/prompt gate;
• material main/head/scope drift now rehydrates inside the same execution window instead of stop and rehydrate handoff;
• URL-only repository entry now resumes safe authorized work instead of forcing numbered next-action choices;
• routine engineering choices/options are not presented by default;
• committed state/README are semantic checkpoints and exclude volatile PR/branch/merge transaction fields;
• live PR/branch/check/merge facts come from GitHub at runtime;
• post-merge state-only reconciliation is explicitly not required;
• AI_CHECKPOINT and M14 authority mirrors distinguish historical authorization provenance from current continuation authority;
• deterministic claims, frontend contract tests, and supervisor audit ratchet these rules.

Static branch consistency pass: all prompt-suppression/state-sync/README/authority/Runner reference checks passed with zero failures.

Required exact-head Code Quality, WorkIntel CI and Windows Certification remain the merge gate.