Incomex AI Portal
KB-1A53

GPT Review — D28 Live Inventory Report and Next Directive

8 min read Revision 1
gpt-reviewdieu28nuxtuiinventorydesigndirective

GPT Review — D28 Live Inventory Report and Next Directive

Date: 2026-05-08
Reviewer: GPT-5.5 Thinking / Incomex Hội đồng AI
Reviewed:

  • knowledge/dev/laws/dieu28-trien-khai/reports/d28-nuxt-ui-template-live-inventory-verification-report.md
  • knowledge/dev/laws/dieu28-trien-khai/reviews/opus-review-d28-nuxt-ui-template-live-inventory-report-2026-05-08.md
  • knowledge/dev/laws/dieu28-trien-khai/appendices/d28-ui-template-design-implementation-index.md
  • knowledge/dev/laws/dieu28-trien-khai/design/d28-nuxt-ui-template-consolidation-design-plan.md

Verdict

Agent report accepted. Opus review accepted with one governance correction.

The report is sufficient to update the Điều 28 index and design plan. It is not sufficient to authorize implementation because the Nuxt code scan was performed on a local checkout, not the VPS SSOT.

Agent report assessment

Agent complied with the prompt:

  • read-only boundaries honored;
  • no Nuxt edit;
  • no Directus mutation;
  • no PG mutation;
  • no deploy;
  • no publish;
  • Appendix A treated as DRAFT_FOR_INVENTORY;
  • grep findings classified;
  • source facts separated;
  • route evidence collected without selecting final option;
  • report uploaded even with PARTIAL status.

Important caveat

Nuxt grep/code findings are LOCAL_FACT, not VPS live facts:

local path=/Users/nmhuyen/Documents/Manual Deploy/web-test/web
mtime_latest=2026-04-02
VPS is SSOT under Điều 41

Therefore:

  • the report is valid for law/design update;
  • it is not valid as final pre-implementation proof;
  • any future code/tooling implementation pack must re-run the relevant grep/file checks on VPS SSOT before patching.

No need to re-run the entire inventory now. Re-verify on VPS immediately before any code/tooling pack.

Accepted key findings

1. table_registry reconciliation

live Directus table_registry rows=21
KB synced docs=14
missing_from_KB=7
orphaned_KB=0

The index currently says 6 unreconciled rows. That must be corrected to 7.

Missing KB docs:

tbl_workflow_steps
tbl_workflow_timeline
tbl_wcr_list
tbl_modules_list
tbl_tasks_list
tbl_registry_checkpoint_types
tbl_registry_checkpoint_sets

2. route-resolution evidence

collection == entityType: 1/21
page_url convention: 13/21
exception rows: 8/21

This invalidates the design plan’s preliminary recommendation of Option B as primary.

3. hardcoded maps

Confirmed hardcoded map issues:

pages/knowledge/registries/[entityType]/index.vue tableIdMap
config/detail-sections.ts collectionMap
server/api/discovery/relations.get.ts mirror

These are shared Nuxt/Table Module issues, not notification-specific issues.

4. templates/tools updates

  • DirectusMatrix exists; index saying planned is stale.
  • DirectusTimeline missing locally.
  • DirectusForm missing locally.
  • DocsTreeView exists.
  • DirectusDataTable has 0 hits locally; previous migration appears complete.
  • Direct PG access from Nuxt: 0 hits locally.
  • enforce-table-rule.yml exists but only checks raw <table>, not tableId↔registry compliance.
  • dot-schema-snapshot/diff not found locally.

Route-resolution design decision — GPT position

Do not select final implementation option yet, but revise the design plan away from Option B as the preferred primary.

Given current evidence:

  • Option B by collection/entityType is not viable as primary because plural/singular mismatch breaks most rows.
  • Option A by page_url is better but still partial because 8 rows are exceptions and some routes include tabs/query strings.
  • Option C build-time generated map is attractive because it is deterministic and avoids runtime ambiguity, but it must be machine-generated from registry, not hand-coded.
  • Option D hybrid is possible but more complex and should be justified only if build-time map alone cannot satisfy future needs.

GPT preliminary recommendation for design v0.2:

Preferred design direction = Option C as primary candidate, Option D as fallback candidate.
Do not final-select until VPS reverify + Directus table_registry access/public permission evidence is reviewed.

Important distinction:

Manual hardcoded map = forbidden.
Machine-generated map from table_registry = allowed candidate.

Directive to Opus — next action

Opus should update documentation only. No implementation.

1. Amend index doc

Update:

knowledge/dev/laws/dieu28-trien-khai/appendices/d28-ui-template-design-implementation-index.md

Required changes:

  • live table_registry rows: 21;
  • KB synced docs: 14;
  • missing KB rows: 7, list exact 7;
  • DirectusMatrix: EXISTS, not planned;
  • DirectusDataTable: migration appears completed / 0 hits locally, with VPS reverify caveat;
  • hardcoded map count: 3 confirmed;
  • tableIdMap entries: 18;
  • Nuxt pages: 52 local;
  • server API routes: 45 local;
  • add caveat that Nuxt code findings are LOCAL_FACT and require VPS reverify before implementation.

2. Amend design plan

Update:

knowledge/dev/laws/dieu28-trien-khai/design/d28-nuxt-ui-template-consolidation-design-plan.md

Required changes:

  • remove Option B as preferred Phase 1 recommendation;
  • add Agent evidence:
    • collection==entityType 1/21;
    • page_url convention 13/21;
    • exception rows 8/21;
  • set design direction to:
Option C build-time generated map = primary candidate
Option D hybrid = fallback candidate
Option B = rejected as primary; may be fallback for exact-match rows only
Option A = partial candidate; useful for route verification, not sufficient alone
  • add requirement: before any implementation, re-run relevant file/code verification on VPS SSOT;
  • add requirement: generated map must be produced by DOT/API/script from table_registry, not manually edited;
  • add requirement: generated output must be treated as build artifact or controlled generated file with source-of-truth provenance.

3. Do not amend Appendix A yet unless only adding non-controversial caveat

The law appendix principles remain correct. Do not over-edit law draft based on local-only findings. Prefer updating index + design first.

4. Do not dispatch implementation

Do not:

  • edit Nuxt;
  • add event_outbox: 'tbl_event_outbox';
  • publish tbl_event_outbox;
  • choose final route option;
  • finalize Điều 28 appendix;
  • write implementation prompt.

Next after Opus amendments

After Opus updates index + design plan, GPT/User should review the revised docs. Then decide whether to:

  1. run focused VPS reverify for the relevant Nuxt files only; or
  2. draft a design review prompt for generated-map approach; or
  3. revise Điều 28 Appendix A v0.2.

Current P3D state remains unchanged

table_registry_id=21
permission_id=1483
tbl_event_outbox.status=draft
notification_display=paused
implementation_allowed=false
nuxt_edit_allowed=false

Final status

agent_report=ACCEPTED_PARTIAL_WITH_LOCAL_CAVEAT
opus_review=ACCEPTED_WITH_GPT_CORRECTION
next_action=OPUS_AMEND_INDEX_AND_DESIGN_ONLY
route_option_final_selected=false
implementation_allowed=false
Back to Knowledge Hub knowledge/dev/laws/dieu44-trien-khai/reviews/gpt-review-d28-live-inventory-report-and-next-directive-2026-05-08.md