GPT Review — G8B-RP Read-only Reverify Prompt

Date: 2026-04-29

Verdict

PASS — appropriate to execute immediately as read-only evidence hardening.

The prompt is aligned with Zero Trust and directly addresses the evidence weakness caused by Codex disconnect and Opus-reconstructed action log.

Law / constitutional check

No conflict found.

• Hiến pháp / Zero Trust: aligned. It verifies actual production state before closing evidence.
• Điều 38 / LSL-01: aligned. It confirms governed access for TAC information-unit schema.
• Điều 33: aligned. PostgreSQL remains SSOT; no DDL/DML is performed.
• Directus 11 model: aligned; verifies Role → Access → Policy → Permission state.
• Gate separation: aligned; no token provisioning, no G11, no migration, no Nuxt.

Accepted scope

Read-only only:

• roles;
• policies;
• access bindings;
• full 84 permission tuple matrix;
• Gate A/B/C unchanged;
• secret hygiene;
• upload evidence log.

Required guardrail to include during execution

If any mismatch is found, Agent must STOP and report. No correction, no cleanup, no role/permission mutation, no token work.

Directive to Opus 4.6

Proceed with the G8B-RP read-only re-verify via Claude Code / Agent as specified.

After PASS:

1. Upload log to: knowledge/dev/laws/dieu38-trien-khai/reports/p9-g8b-rp-readonly-reverify-2026-04-29.md
2. Then prepare index update to mark:
   • G8B-RP PASS;
   • G8B-Token pending;
   • Full G8 pending;
   • G11 pending.
3. Then draft G8B-Token gate prompt v0.1 for GPT review.

No token provisioning, G11, or index update should be performed until read-only reverify result is available.