Bible v19.1.10 + Salesforce v7.1.0 merged to main on 2026-05-06
Decision
Bible v19.1.10 (R48 Complete-Capability Invariant + Check 15 Symmetric Metadata Reader-Writer Audit) and Salesforce MCP v7.1.0 (9 new tools, 673 to 682, R48 pilot) both merged to main on 2026-05-06 after 12 Codex hard-gate review rounds total + 2 post-merge Pristine Sweep cleanups. The 19-permission System Administrator Clone tightening is now ONE tool call. Skill v5.1 paired-synced with byte-identical mirrors and new references/profile-complete.md.
Rationale
AJ directive 2026-05-05 zero-limitation policy + 2026-05-06 course-correction “full deliver and merge everything into main, zero compromises” drove deliver-not-stage. Codex hard-gate caught 27 real findings across 12 review rounds; each fix independently verified before commit. Pre-deploy audit: 2 of 3 items verified in-session via source inspection (mdapi unpackaged kwarg shape, MutingPermissionSet folder casing internal consistency); 1 deferred (bundled JSON edge-direction audit) with empirical-bootstrap mitigation in permission_deps.py. Production cutover (Phase 9 fullcopy + production smoke against the 19-permission scenario) is AJ-driven post-merge.
Alternatives Rejected
Defer Salesforce branch to next session (rejected — AJ explicit “delivery completion here”); rc-sequenced PRs over 4-6 sessions (rejected — AJ course-corrected); override Codex hard-gate via log_and_alert flag (rejected — gate caught real bugs every round, bypassing would have shipped silent-no-op + mdapi signature-mismatch + closure-direction bugs to production).
Outcome
Pending