Macro-5 Codex Review Packet — R2-B2 (2026-06-19)
Macro-5 Codex Review Packet — R2-B2 (2026-06-19)
Date: 2026-06-19 · Workstream: R2-B2-MACRO-5-STAGING-BUILD-AUTHORIZATION-PACKAGE-2026-06-19 (Deliverable 94 of 110) · Editorial revision: rev1
Class: Codex review packet · READ-ONLY · NON-ENACTING · NON-AUTHORIZING · NO write performed.
Metadata convention. Editorial revision (rev1) only. AgentData storage revision/
content_lengthauthoritative at read time.
0. Status and non-authorization
STATUS: PASS — engineering / design-only. The adversarial surface Codex should attack across the 110 files. Engineering PASS ≠ authority PASS. Default: HOLD.
1. Purpose
Give Codex the exact checks to refute, with expected verdicts.
2. Sources / evidence read
Macro-4 Codex review packet (D73) + adversarial attack list (D89); prompt §2 (done criteria) + §7 (self-check). Main process, no reader-agents.
3. Accepted baseline (carried)
Codex is the adversary of record; the macro self-checks but does not certify itself.
4. Evidence / analysis — attack surface (full list in Deliverable 95)
| # | Attack | Where |
|---|---|---|
| MX-1 | any executable SQL/DDL/migration/rollback script present? | 13–22, 84 |
| MX-2 | any staging schema/table/corpus created? | 9, 54, 84 |
| MX-3 | actual B2 TD opened; entry gate honestly NO-GO? | 81 |
| MX-4 | B2 output beyond candidate-only inspect_*? |
25, 32, 34 |
| MX-5 | channel selected as runtime authority? | 47 |
| MX-6 | S2 owner assigned / ownership row written? | 30, 48 |
| MX-7 | Điều 0-G adopted/recovered/patched? | 49, 69 |
| MX-8 | bad-input test run / digest produced? | 72–77 |
| MX-9 | IO contract = mega-registry / shared surface? | 15, 88, 91 |
| MX-10 | any production write (preflight/postflight/delete-fast)? | 41–71, 105 |
| MX-11 | B5/B7/R1 scope creep? | 86, 87 |
| MX-12 | v0.1 overwritten / v0.2 promoted? | 92 |
| MX-13 | any blocker falsely resolved? | 100 |
| MX-14 | any deliverable not independently discardable (NOT_LEGO)? | 3 |
| MX-15 | mega-birth pipeline (fused B1+B2+B3+B4)? | 90 |
| MX-16 | engineering PASS used as authority? | 93, 82 |
| MX-17 | reader-agents / local-prose inference used? | 105, 110 |
5. Candidate / requirement / gate / result
Codex should return a per-MX verdict + any new caveat. Self-checked expectation: MX-1…MX-17 not triggered.
6. Owner-gated future work
Codex review is a control step; it authorizes nothing.
7. What remains unresolved
Codex verdict pending.
8. Ready for GPT/Codex review
Yes — this packet is the entry point.