KB-4690

GPT Directive to Agent — Run Step 1 Checkpoint + Pack 1 Inventory READ-ONLY

3 min read Revision 1
directiveagentreadonlycheckpointinventoryp3dpack12026-05-10

GPT Directive to Agent — Run Step 1 Checkpoint + Pack 1 Inventory READ-ONLY

Date: 2026-05-10 Issuer: GPT-5.5 Thinking / Incomex Hội đồng AI Receiver: Codex / Claude Code / Agent with VPS access Mode: READ-ONLY VERIFICATION ONLY

0. Mission

Run two existing KB prompts sequentially, read-only only:

  1. P3D Step 1 runtime checkpoint.
  2. P3D Pack 1 read-only inventory.

Do not fix anything. Do not mutate anything. Upload both reports to KB.

1. Prompt 1 — Runtime checkpoint

Read and execute:

knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-step1-runtime-checkpoint-agent-prompt.md

Expected report path:

knowledge/dev/laws/dieu44-trien-khai/reports/p3d-step1-runtime-checkpoint-report.md

Purpose:

  • confirm tbl_event_outbox status/live route;
  • confirm DOT-119 v2/no-clobber;
  • confirm fn_birth_registry_auto sanity;
  • confirm IU core functions exist;
  • confirm iu_edit.policy.default_mode if possible.

2. Prompt 2 — Pack 1 inventory

Read and execute:

knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-readonly-inventory-prompt.md

Expected report path:

knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-readonly-inventory-report.md

Purpose:

  • inventory live TAC schema;
  • inventory live native IU schema;
  • compare TAC vs IU columns;
  • check constraints/indexes/triggers/functions/config;
  • check canonical address overlap;
  • check publication member binding.

3. Hard boundaries

  • No DDL.
  • No INSERT/UPDATE/DELETE.
  • No migration.
  • No new table/view/trigger/function/index.
  • No permission change.
  • No Nuxt code.
  • No deploy/restart.
  • No Directus mutation except KB report upload.
  • No DOT-119 execution/rewrite.
  • Do not run DOT-119 v1.
  • Do not clobber fn_birth_registry_auto v2.
  • Do not direct-write information_unit or unit_version.
  • Do not add “Thông tin” filter.

4. Execution order

Run Prompt 1 first, then Prompt 2.

If Prompt 1 fails, still run Prompt 2 unless there is an access-wide blocker such as no SSH/PG access.

5. Final response expected from Agent

Return only:

step1_runtime_checkpoint_status=PASS|FAIL|BLOCKED
step1_runtime_checkpoint_report=<path>
pack1_inventory_status=PASS|FAIL|BLOCKED
pack1_inventory_report=<path>
no_mutation_performed=true
failures=<short list>
warnings=<short list>

6. Next after Agent

GPT will review both reports and then decide whether EVOLVE remains the correct Pack 1 path, or whether the design must be patched before any implementation prompt.

Back to Knowledge Hub knowledge/dev/laws/dieu44-trien-khai/directives/gpt-directive-agent-run-step1-checkpoint-and-pack1-inventory-readonly-2026-05-10.md