GPT Operating Directive to Opus — MVP Information Unit Next Steps

Date: 2026-04-29

User direction

User agrees with the current roadmap and wants to complete the MVP information-unit workflow first. After the MVP is stable, User will provide additional product requirements to move toward production.

GPT is delegated to operate and approve intermediate technical work. Do not ask User to review technical prompts/reports unless a business/product decision is required.

Current accepted state

• P9 foundation: PASS.
• P10A technical pipeline: PASS.
• P10A clean human-readable render: PASS expected/under final completion flow.
• Roadmap requirement recorded:
  • /knowledge/laws should become PG-assembled document reader later;
  • Tree View;
  • Reader Mode / Review Mode / Debug Mode;
  • permission-controlled unit-boundary overlay;
  • auto numbering/reorder/split/merge architecture.

Strategic priority

Complete MVP Information Unit before production product expansion.

MVP means:

1. A document can be segmented into information units.
2. Units are stored in PG.
3. Units can be reconstructed into a clean human-readable document.
4. The structure/order/tree can be verified.
5. The process can be repeated for more than one document.
6. Basic future requirements are captured without prematurely implementing them.

Immediate sequence

Step 1 — Finish/confirm P10A-2C

If P10A-2C deliverables are already uploaded, produce concise GPT review:

• confirm clean render path;
• confirm no substantive content drift;
• confirm final user-facing file path;
• confirm P10A COMPLETE.

Do not ask User to inspect technical reports. Only provide the final clean file path and summary.

Step 2 — P10B Multi-document validation

Draft a concise P10B plan for 2 more documents.

Purpose: prove that the MVP works beyond Điều 35.

Constraints:

• do not broaden into production migration;
• no Nuxt implementation yet;
• no KG/vector;
• no schema redesign unless a blocker appears;
• keep documents mutable/proposed/draft;
• produce clean human-readable reconstructed outputs for each document;
• compare against originals;
• record ordering/tree issues.

Candidate documents should be selected to test variety, e.g. one short law and one structurally different document/SOP/design note.

P10B output should include:

• original document;
• reconstructed clean document;
• diff/readability report;
• tree/outline report;
• row-count/report;
• summary of segmentation/routing issues.

Step 3 — P10D design note only, not implementation

In parallel or immediately after P10B plan, prepare a short P10D product/design note:

P10D — Nuxt Laws Page Tree View + Reader/Review Modes + Auto Numbering Design

P10D must cover:

• reuse /knowledge/laws route and homepage tab;
• left tree view, right document reader;
• Reader Mode clean by default;
• Review Mode overlay toggle visible only to authorized roles;
• Debug Mode optional;
• auto numbering/reorder/split/merge requirements;
• stable ID vs structural position vs human-facing numbering;
• read-only first implementation slice.

Do not implement P10D until User/GPT chooses after MVP validation.

Important product requirement to carry forward

The unit-boundary overlay is a permission-controlled display layer:

• OFF = clean document;
• ON = subtle unit boundaries for authorized reviewers/admins;
• content in PG must not be modified to display boundaries;
• access controlled by admin/permissions.

Operating style

• Keep Opus outputs concise.
• Avoid overly granular prompts unless there is production mutation.
• GPT handles technical approvals.
• User should see product-visible outcomes, not technical artifacts.
• Do not ask User to approve technical details unless there is a business/product decision.

Directive

Opus 4.6 should now proceed as follows:

1. If P10A-2C has not been run, dispatch/finish P10A-2C and return only result summary + clean file path.
2. If P10A-2C is complete, produce P10A final closeout and draft P10B multi-document validation plan.
3. Include the P10D roadmap note as future product/design gate, but do not start implementation.

Stop after producing the next plan/report for GPT review.