Turn conversations into specs, and specs into a working service.
Turn your requirements conversation with Codex into a reviewable Markdown contract, then in the Antigravity IDE build
a mobile customer ordering screen and a PC admin screen based on it, and verify both the design and the service.
Rebuild the conversation into project requirements
Split the conversation's requests into facts, assumptions, questions and decisions, and set the build scope and success criteria yourself.
25POINTS
Input Requirements conversation logOutput PROJECT_BRIEF · DECISION_LOGDone signal Assumptions separated from final decisions
01
Write a one-sentence definition of the service yourself
Starting from "Customers order sandwiches and drinks on their phones, and the admin manages menu, stock and order status on a PC," redefine the target users, problem to solve, key actions and out-of-scope items in your own words.
💡 Student-first rule: Before opening Codex, always write your own draft (users·key flow·out of scope) and 5 questions you can't answer yet.
02
Ask Codex to structure requirements and find counter-examples
Set limits so AI can't invent features, and have it surface missing exceptions and data questions.
CODEX STRUCTURE PROMPT
Structure the following conversation into a draft service spec.
Don't invent new facts; clearly label [Fact], [Assumption], [Question] and [Decision].
Separate the requirements for the customer mobile ordering screen and the PC admin screen,
and suggest missing exceptions, data consistency and verification questions with priorities.
03
The student, not AI, must own the decisions
Separate facts stated in the conversation from AI guesses, and leave evidence and review criteria for every final decision.
FACTA requirement stated in the conversationASSUMPTIONA guess by AI or the teamQUESTIONSomething that needs checkingDECISIONReason · owner · criteria
04
Fix separate responsibilities for the four top-level folders
Planning, analysis, building and verification are independent areas with different responsibilities. Keep the folders' roles from mixing.
Fix data contracts and requirement traceability before screens
Define a unique ID, category, base price, option groups, sale status and stock count for 10+ sample menu items, and link everything from requirements to tests by ID.
Take only the reference's UX principles; don't copy its assets
Use fresh-ingredient colors (Fresh Route), the fun of step-by-step option picking and clear progress status as reference, but never copy any outside brand's logos, text or images.
04
Cross-audit your docs with Codex for contradictions and gaps
CROSS-DOC AUDIT PROMPT
Cross-review the planning·analysis·verification docs below.
Don't add new requirements; only find these:
1. Mismatched terms · IDs · order status values between docs
2. Features in screen specs that are missing data/API definitions
3. Vague done criteria that can't be observed or measured
4. Risks around personal data · payments · permissions · copyright, and questions a person must decide
PASS GATE
Mission 2 done checklist
0 / 8
MISSION 03 · BUILD
Build in the Antigravity IDE following the doc contract
Use the approved Markdown specs as the only basis for building, and build the mobile customer and PC admin screens in vertical cycles.
30POINTS
Input Approved planning·analysis·design MDOutput Working service (2 Views) · AI_WORK_LOGDone signal Live mobile·PC sync demo
01
Declare a docs-first build contract to Antigravity
Instead of asking for the whole app at once, lock the scope per requirement ID and repeat build·verify.
ANTIGRAVITY BUILD CONTRACT
Use only the approved MD docs in service-planning and service-analysis as the basis for building.
This build's scope is one item: [REQ-___ / screen ID].
Before writing code, report the related doc paths, files to change and anything unclear,
and after building, summarize the changed files, how to run it and results per acceptance criterion.
02
Finish in three vertical cycles (Vertical Slices)
Cycle 1 · Customer mobile flowBrowse menu → Details·pick options → Add to cart·adjust quantity → Order done and check order number (mobile 360px~430px)
Cycle 2 · Admin PC flowLive new order alerts·list → View order details → Change status (received→cooking→ready for pickup→done) → Sold-out·menu management (PC 1280px+)
Cycle 3 · Exception and recovery flowPrompt for missing required options, block sold-out items, prevent empty cart orders, retry recovery on network/save failure
03
Run the real integrated demo service to compare and check
This workspace includes a live-synced customer mobile + PC admin service built from the doc contract and Fresh Route design concept.
If the build must differ from the docs, update the docs first
Don't just patch code for changes found while building; update the decision record (DEC) and spec together to keep docs and code consistent.
Check doc ID→Build minimum scope→Run and test→Work log (AI_WORK_LOG)→Next requirement ID
PASS GATE
Mission 3 done checklist
0 / 8
MISSION 04 · VERIFY
Verify the design and the service against different criteria
Check doc completeness (design verification) separately from the real app's behavior·usability (service verification), and record the gaps you find as defects and a learning retrospective.
20POINTS
Input Specs · Built service · Work logOutput Design·service verification report · RetroDone signal Evidence-based pass (0 critical defects)
01
Design Validation checks consistency between docs
Do terms, requirement IDs, order status values and data field names match in every doc?
Does every key requirement have observable Acceptance Criteria and a linked test scenario?
Does the data flow continue unbroken from creating an order on customer mobile to changing status on PC admin?
Are permission boundaries, personal data exclusion rules and exception/failure conditions all covered?
02
Service Validation checks real run evidence
Actually reproduce the main user flows at mobile (360px~430px) and PC (1280px+) sizes, and capture·record them.
Test not only normal orders but also empty carts, missing required options, blocked sold-out items and cancelled/rejected states.
Check that base price + option extras × quantity = final payment amount, and that order status changes match on both screens.
Check keyboard navigation, button labels, color contrast (Fresh Route accessibility) and loading/toast feedback.
03
Attach reproduction steps and post-fix re-check evidence to every defect
DEFECT REPORT TEMPLATE
## DEF-001 — [defect title]
- Linked requirement / test ID: REQ-___ / TEST-___
- Test environment and reproduction steps:
- Expected result (per docs):
- Actual result (when found):
- Screenshot · console log · related files:
- Severity (Critical/High/Medium/Low) and status:
- Cause and fix:
- Re-check result after fix:
04
Reflect on the differences between AI suggestions and human final decisions
Summarize, with IDs, what Codex and Antigravity structured·built quickly, what AI missed or over-assumed, the design·code decisions a person stepped in to change, and the first question you'll check in your next project.
PASS GATE
Mission 4 done checklist
0 / 8
EDITOR WORKSPACE · 4 CORE DOMAINS
Project work files and run environment
The class page and the real sample project (planning·analysis·build·verification) are in the same folder. Click a card to open a doc or run the built app.
Separate facts · assumptions · questions · decisions
7 pts
MVP scope·success/failure criteria
7 pts
Reasons for adopting·changing·rejecting AI suggestions
6 pts
02 · 25 pts
Markdown spec package
Mobile·PC screen·state specs
7 pts
Menu data·API·analytics contract
7 pts
Design principles·safety/copyright boundaries
5 pts
REQ→UI→DATA→API→TEST traceability
6 pts
03 · 30 pts
Antigravity service build
Key customer mobile ordering flow
8 pts
PC admin order checking·status changes
7 pts
Exceptions·recovery·totals·status consistency
6 pts
Matches approved doc IDs
5 pts
AI work log·human judgment records
4 pts
04 · 20 pts
Design·service verification and retro
Design verification (doc consistency·gaps)
5 pts
Service verification (reproduce mobile·PC·exceptions)
6 pts
Defect records and re-check after fixes
5 pts
AI vs human judgment retro·presentation clarity
4 pts
Final pass criteria70+ total points (50%+ on each mission)All 4 mission outputs submittedKey customer mobile + PC admin flows run0 Critical defectsA person verifies·explains AI output directly
⚠️ Safety·ethics and AI accountability rules you must follow (violation = automatic fail)
Never store real personal data (names·phone numbers·addresses), real payment info or secret API keys in code or analytics events.
Don't copy logos, signature text, product images or source code from commercial services (like Subway); use your own design concept (Fresh Route) and sample menu data.
Write your own draft before opening AI, and don't let AI add requirements that weren't in the conversation as if they were facts.
Record at least 2 cases where you rejected or changed AI suggestions, and explain your architecture and verification evidence in your own words in the final presentation.
SAVE YOUR WORK
Save your current progress as a Markdown file
Checklist progress is saved in your browser automatically, and you can download it right away as a progress log for submission (DOC_DRIVEN_PBL_PROGRESS.md).