📊 Summary/diagram view ← To list
Day 1 Codex AI Requirements Definition Environmental setting & practice mission

Codex-based requirements definition training mission

You will learn the entire planning workflow from performing summarization (preprocessing) from informal meeting minutes conversations, defining functional specifications and exception flows, building decision questionnaires, and integrating and sharing final documents by email.

Select progression steps

Step 0

Conversation summary preprocessing

Extract key summary notes from original meeting transcripts

Step 1

Meeting notes analysis

Sort users, functions, and business policies from meeting minutes summary

Step 2

Define functional requirements

Convert extracted feature candidates into testable formal specification sentences

Step 3

Normal and exception flows

Supplement details of user scenarios and exception cases (failure, duplication, etc.)

Step 4

Undecided Planning Questions

Turn undecided issues into a list of stakeholder verification questions

Step 5

Document consolidation and email sharing

Complete document through AI Lint verification and send via Gmail plugin link

Mission 1

Pet care matching

[Assignment] Creation of planning document based on pet sitter matching service meeting minutes

Mission 2

Unmanned Study Room & IoT

[Task] Real-time space reservation and smart device control specification

Step 0 Deriving key summaries from detailed conversation transcripts (preprocessing)

📥 Step input (Input)

Original minutes of the first meeting in the form of an irregular ping-pong conversation (meeting-notes-raw.md)

📤 Step output (Output)

Key Summary Memo File (meeting-notes.md)

🛠️ Detailed guide to follow in practice

  1. Secure original meeting minutes: Click the download button below to download the original transcript of the first meeting. meeting-notes-raw.md Download the file and place it in your working folder or generate the file contents.
  2. Pretreatment instructions: Open the Codex/ChatGPT dialog box you want to use and click the button below. Codex request example Send a prompt.
  3. Save results: AI copies the refined summary memo text extracted from the raw Ping Pong conversation meeting-notes.md Save it as a file. (This file will be the input data for the next step 1 exercise.)

mission situation

The original meeting minutes written at the planning meeting are a mixture of ping-pong conversations and casual chats between each team member (representative, planner, developer, and designer), making it too distracting to directly extract and analyze requirements. Learners should use Codex as a preprocessing tool to derive planning memos from meeting minutes that summarize only key business requirements.

Learner Mission

The original transcript of the first meeting: meeting-notes-raw.md Write to file or download.

meeting-notes-raw.md original text (scroll)# 1st kick-off meeting minutes to establish expert consultation service

**Date**: 2026-07-10 14:00 - 15:30
**Attendees**: CEO, planner (PM), developer (Dev), designer (Designer)

---

**Representative**: Hello everyone. Today’s meeting is to discuss the basic requirements and establish direction for our newly launched “Online Expert Consultation Service.” The key is the process where users come to our platform, find experts in various fields, make reservations for consultations, and complete payment smoothly.

**planning person**: Yes, CEO. Let’s start our discussion with the basic sign-up process for the service. Does the structure require users to sign up to use the service?

**Representative**: Yes, that is correct. Users must first sign up to be able to smoothly use functions such as expert search or reservation. Non-member status only allows for simple browsing.

**Designer**: After signing up, you will be searching for experts. What are your thoughts on the expert profile page and classification criteria?

**Representative**: Experts must first be categorized by field. For example, there might be categories such as IT consulting, legal consulting, and design coaching. Users should be able to quickly find the expert they want by selecting a category or directly entering a search term.

**Designer**: Then, a list of experts will appear in the search results, and you will then go to the details page and select a schedule. How do I choose the date and time?

**Representative**: Yes, when the user selects a specific expert, the schedule for consultation with that expert should be displayed in the form of a calendar or timeline. Users must be able to freely select the date and time they wish to receive consultation and apply for a reservation.

**Developer**: How does the payment process work immediately after requesting a reservation? Does payment have to be made immediately upon requesting a reservation to confirm the reservation?

**Representative**: The final reservation must move to “confirmed” status once payment has been completed. If payment fails due to exceeding the PG company's limit or for other reasons, the reservation should not be processed as done, but should be stuck in waiting or cancellation status.

**Developer**: This is a business rule that states that when payment fails, the reservation should not be created or the lock should be released. Where can professionals check their scheduled schedules?

**Representative**: Experts must have an expert-only dashboard within our system, where they can monitor and manage their reservation request details and confirmed schedules in real time.

**planning person**: Signing up as an expert will be different from regular signing up. Is there a separate authentication or approval process required? You can't have just anyone act as an expert.

**Representative**: That's a good point. Experts cannot work immediately after applying for membership, but the site administrator must review the submitted documents and qualifications and "approve" them before they can be exposed to the expert list and participate in the reservation system, so we will set the administrator approval policy as required.

**planner**: When a reservation is made, a notification should be sent to the user or expert to improve development convenience and usability. Which notification channel do you prefer?

**Representative**: Notifications must be sent unconditionally. We need to notify you when your reservation is confirmed or cancelled. However, I think we need to consider development resources and cost issues more closely as to whether to send notifications by email or text message to a mobile phone. Let's not confirm this part yet.

**Developer**: Do we need to set rules for cancellation fees and refunds now?

**Representative**: The cancellation fee policy is also something that requires legal review and consultation with the marketing department. It's difficult to decide today, so let's leave this as a pending matter that hasn't been decided yet. It will be confirmed at the second meeting later.

**planner**: Okay. We will roughly compile the policies and pending matters decided today and organize them into a summary memo of the pre-processing meeting for the draft specification.

Codex request example (preprocessing prompt)

PromptRead the original meeting minutes from meeting-notes-raw.md, remove unnecessary conversation context and context between speakers, and
Write down information about service users, functional requirements, system policies, and pending matters in the form of a compact, unstructured summary memo of approximately 15 lines.
Please do not arbitrarily add fictitious content that is not in the original text.

🤖 Codex role at this stage

Summarizing and preprocessing tool (Summarizer): Remove unnecessary remarks and speaker context from the conversation and extract key summary notes required for precise design.

Step 1 Find key takeaways from meeting notes

📥 Step input (Input)

Summary planning memo file refined in step 0 (meeting-notes.md)

📤 Step output (Output)

  • • List of service user roles.
  • • Feature candidate list
  • • List of business rules
  • • List of pending matters

🛠️ Detailed guide to follow in practice

  1. Prepare meeting notes: Step 0 Summary meeting-notes.md Place the file. You can directly download the refined example file by clicking the download button below.
  2. AI analysis request: Following the step 0 conversation, below Codex request example Send a prompt.
  3. classification review: Derived user list (Customer, Expert, Admin) and confirmed/undecided policy classification Learner Checklist See if it meets the criteria.

mission situation

It is time to analyze the level 0 planning summary notes (`meeting-notes.md`) refined from meeting conversations between representatives, planners, developers, etc. Based on this summary, learners must accurately categorize the main objects of the service (user roles, functions, business rules, and policies).

Learner Mission

refined meeting-notes.md Prepare a summary file or use the example download below.

meeting-notes.md summary (scroll)We are planning to create an online expert consultation service.

Users can search for experts after signing up.
Experts are classified by field.
Users can select the consultation date and time.
Once payment is completed, the reservation is confirmed.
If payment fails, the reservation cannot be completed.

Experts can confirm their reservations.
Administrators can approve expert membership.

A notification should be sent when the reservation is completed.
It has not yet been decided whether notifications will be sent by email or text.
The cancellation fee policy has not yet been determined.

Codex request example

PromptRead meeting-notes.md and identify the following:

1. Service users
2. Functions performed by the user
3. Functions performed by the system
4. Business rules
5. Matters not yet decided

Do not add anything that is not in the meeting notes.
Please indicate unclear information separately.

Don't edit the file yet, just show the analysis results.

Learner Checklist

  • Are users classified into Customer, Expert, and Admin?
  • Has the relationship between payment success and reservation confirmation been confirmed?
  • Are notification methods and cancellation policies separated as pending matters?
  • Didn't Codex add a feature that wasn't in the meeting notes?

🤖 Codex role at this stage

Analysis Tool (Analyst): Logically identify user roles, specification candidates, and business rules from a refined planning summary memo.

Step 2 Creating functional requirements statements

📥 Step input (Input)

Step 1 output (list of user roles, list of feature candidates, list of business rules, list of pending decisions)

📤 Step output (Output)

functional-requirements.md (FR-001 ~ FR-004 standardized functional requirements definition table)

🛠️ Detailed guide to follow in practice

  1. Create requirements template: In the working folder functional-requirements.md Create the file below and Learner MissionCopy and paste the table template.
  2. send prompt: Following the first stage conversation, below Codex request example Copy the prompt and send it to the chat window.
  3. Copy & save results: Copy the specification sentence in tabular format (Markdown) generated by AI. functional-requirements.mdPaste it in and save it.
  4. Ambiguity Filtering: If the completed sentence contains ambiguous modifiers such as “conveniently” or “quickly,” try asking the AI to correct the quantification again.

mission situation

In step 1, we found the feature candidates needed for the service. In this step, we turn the feature candidates into actual requirements statements. It should be written in sentences that developers and planners can equally understand, rather than simple function names.

Criteria for writing requirements

[User or system] must be able to perform [function or action] under [specific conditions].

Learner Mission

Change the feature candidates extracted in step 1 to the following format:

Requirement ID user Function name Requirements Description
FR-001 Customer Sign up (blank space to be filled by the learner)
FR-002 Customer expert inquiry
FR-003 Customer Select consultation time
FR-004 System Reservation confirmed

Codex request example

PromptWrite the feature candidates extracted in step 1 as functional requirements.

Apply the following rules:

1. Requirement IDs are assigned in order starting from FR-001.
2. Write only one function for one requirement.
3. Write clearly who the user is.
4. Includes the conditions under which the function is executed.
5. Write sentences that can be tested.
6. Do not use ambiguous expressions such as “conveniently,” “appropriately,” or “quickly.”
7. Do not add anything that is not in the meeting notes.

Write the results in a Markdown table.

🤖 Codex role at this stage

Technical requirements writing tool (Technical Writer): Produce quantitative and verifiable sentences based on templates and formal formulas.

Step 3 Adding normal and exception flows

📥 Step input (Input)

Step 2 output (functional-requirements.md table)

📤 Step output (Output)

  • • Normal/exceptional flow supplementary design
  • • List of requirements verification criteria
  • • List of policies that require additional confirmation

🛠️ Detailed guide to follow in practice

  1. Selection of requirements: Subject to supplementation FR-004 Reservation confirmed Find an item in a file.
  2. Run prompt: Following the second step dialogue window, Codex request example Copy and type the prompt.
  3. Flow complement reflection: Detailed normal flow and exception flow returned by AI, verification criteria text block functional-requirements.mdPaste it further under the corresponding ID and save the file.
  4. Random creation filtering: Check whether the AI has imagined an exception policy that is not in the original text and left it as “confirmation required” rather than arbitrarily defining it.

mission situation

We have written a requirements statement, but only the situations under which it operates normally are defined. In actual services, exception situations such as payment failure, no data, and duplicate reservations frequently occur. In this step, we add normal and exception flows to each requirement.

Learner Mission

Select the following features from the functional requirements in step 2 and define normal/exceptional flows:

  • expert inquiry
  • Select consultation time
  • Payment and reservation confirmation
Template### FR-004 Reservation confirmed

- User:
- Prerequisites:
- Normal flow:
- Exception flow:
- Verification criteria:

Codex request example

PromptFor each requirement in functional-requirements.md
Add normal flow and exception flow.

Please apply the following criteria.

1. The normal flow is written step by step from the user’s initial action to the completion result.
2. The exception flow reviews failure, duplicate, no data, and unauthorized situations.
3. Do not arbitrarily decide on exception policies that cannot be confirmed in the meeting notes.
4. Contents that require policy decisions are marked as “confirmation required.”
5. Write at least one verification criterion for each requirement.

🤖 Codex role at this stage

Exception Exploration Tool (QC Explorer): Review and propose countermeasures for design flaws and abnormal input errors (duplication, timeout, exception handling).

Step 4 Create pending decisions and planning questions

📥 Step input (Input)

Among the 3rd step outputs, ‘List of policies and pending matters that require additional confirmation’

📤 Step output (Output)

open-questions.md (Decision Planning Question Table)

🛠️ Detailed guide to follow in practice

  1. Create a planning questionnaire: In the working folder open-questions.md Create a file.
  2. send prompt: Following the 3rd step conversation, below Codex request example Sends a prompt to print a table of pending decisions.
  3. Question history: Table organized by AI open-questions.md Paste it into a file and save it.
  4. role monitoring: Check whether the AI has arbitrarily answered the question directly and does not violate the purpose of the question (maintaining the question state for decision-making).

mission situation

In the process of specifying requirements, business policies that could not be determined immediately were discovered. These pending decisions must be converted into specific, systematic questions to receive planning feedback from stakeholders and decision makers.

Codex request example

PromptBy reviewing the current requirements document
Find unconfirmed or ambiguous content.

Each item can be confirmed by the planner to stakeholders.
Please write in question form.

Please include the following items:

1. Question ID
2. Related Requirement ID
3. Questions to ask
4. Why this decision is necessary
5. Functions affected when decisions are not made
6. Priorities
7. Status

Please don't make up answers to questions arbitrarily.

🤖 Codex role at this stage

Decision-making facilitation tool (Question Maker): Divide the functions and importance affected by the withholding policy and list the issue questions required for agreement in the format.

Step 5 Create requirements workbook and share via email (Gmail plugin)

📥 Step input (Input)

  • • meeting-notes.md etc. specification files
  • • **Gmail Plugin** integration within Codex Client

📤 Step output (Output)

  • • requirements-specification.md (Comprehensive version)
  • • Completed actual email sending through Gmail plugin API

🔌 [Core Preparation] Codex Client Gmail plugin setup and integration

  1. at the top right of the Codex Client screen. [Plugins (Plugin Store)] Click the icon.
  2. from the list `Gmail` Find the plugin and press the **[Install]** button.
  3. Once installation is complete, the Google authentication window will appear. **Log in with your Google/Gmail account** and complete **OAuth email permission approval** to allow the AI ​​agent to send emails.

🛠️ Detailed guide to follow in practice

  1. Document merging and primary verification: Send the prompt for **Learner Mission 1 (Integration and AI Verification)** below to the AI dialog window to create an integrated requirements definition. requirements-specification.md Request creation.
  2. Ambiguity correction feedback: In the error result window detected by AI WARNING If there are any warning words, we share feedback with AI to refine the specification and build it into a final approved version.
  3. Gmail plugin email sending instructions: To utilize the linked plugin, copy the **Learner Mission 2 (Send Gmail plugin)** prompt and send it to the AI dialog window.
  4. Shipping confirmation: Open the mailbox of the recipient's email address or open the 'Sent mailbox' of your own email to verify that the email was actually sent successfully.

Learner Mission 1 (Integration and AI Validation Prompt)

Prompt 1Read the following three files and create a requirements-specification.md file, which is a requirements definition that merges them.

-meeting-notes.md
- functional-requirements.md
- open-questions.md

At the same time, analyze errors in the requirements document based on the six rules below.

[Verification Rules]
1. Requirement ID duplicate?
2. Is the requirement ID missing?
3. Is the user role missing?
4. Whether normal and exception flows are missing
5. Whether verification criteria are omitted
6. Whether to use ambiguous expressions (ambiguous words: conveniently, quickly, as necessary, appropriately, normally)

The error verification results are displayed at the beginning of the output window in the format below, and the contents of the requirements-specification.md file are combined and displayed below.

[Output Format]
PASS: No duplicate requirement IDs
ERROR: FR-004 Verification criteria missing
WARNING: Use of ambiguous expression “conveniently” in FR-007

Learner Mission 2 (Gmail Plugin Send Prompt)

Prompt 2Using the Gmail plugin, the contents of the requirements-specification.md file you just verified and
Send an email according to the conditions below based on the list of open questions in open-questions.md.

- Recipient (To): [Enter the email address of the representative or decision maker who will receive feedback here]
- Email subject: [Approval Request] Review of expert consultation service requirements definition and coordination of undetermined policies
- Body structure:
  - Report on completion of integrated verification of planning documents
  - Major open questions requiring urgent planning decisions (Summary tips for Q-001 ~ Q-004)
  - Feedback response deadline information

[Alternative solution] Local mail client and manual sending

If sending the plugin is difficult due to Google account security restrictions, you can write or copy the email by clicking the button below.

🤖 Codex role at this stage

Automation and plugin integration tool (Plugin Integrator): After completing the internal document analysis stage, remotely link the Gmail plug-in API to send business agency communication emails.

Practice task 1 Pet home care and matching service

📥 Assignment input (Input)

  • • pet-care-meeting-notes.md meeting minutes
  • • Codex writing prompts learned in previous exercises in steps 1-5.

📤 Assignment output

  • • pet-care-requirements.md (Functional and Exception Flow Requirements Definition)
  • • Email approval request for undetermined policy coordination

💡 Mission background and task goals

These are the planning meeting minutes of a platform that connects pet owners (customers) with professional pet sitters. The goal is to clearly specify the guardian's identity and pet information registration, pet sitter's qualification screening and manager approval, payment deadline restrictions after matching is accepted, real-time change of care status process, and exception handling logic in case of service cancellation and pet sitter no-show.

🛠️ Self-practice guide

  1. Download meeting minutes: Click the button below pet-care-meeting-notes.md Download the original file.
  2. Requirements Documentation: Based on the downloaded meeting minutes, Codex AI is instructed to specify in detail user roles (general user, pet sitter, administrator), functions (signup, search, reservation, payment, status change), and exception flows (payment timeout, cancellation penalty, no-show response).
  3. AI Lint Verification: After writing the plan, the requirements document is closely inspected by AI using five-step verification rules (ID duplication/omission, normal/exception flow omission, ambiguous expression check) and the final requirements definition is built into an approved version.

Learner self-practice materials

Click the button below to download the original meeting minutes to complete Mission 1.

Recommended AI Practice Prompts

Mission 1 Prompt[Role: IT Professional Business Analyst]
Analyze the pet-care-meeting-notes.md file you just downloaded and write a requirements definition, pet-care-requirements.md, in the following format.

1. Overview and service structure
2. Define user roles (guardian, pet sitter, administrator)
3. List of core functional requirements (ID format: PCR-001, PCR-002...)
4. Normal flow and exception flow (e.g. processing of payment expiration exceeding 1 hour, 200% point compensation for pet sitter no-show, replacement pet sitter recommendation process, etc.)
5. List of non-decision questions

Specify it to the level that is as detailed and developmentable as possible, and apply the AI Lint verification conditions used in step 5 and show the self-verification results at the beginning.

🎯 Key evaluation and verification criteria

• Are the conditions for ‘administrator approval’ clearly specified after registering as a pet sitter?
• Have deadline exceptions such as ‘cancellation if payment is not made within 1 hour’ been reflected after applying for matching?
• Are compensation policies for pet sitter no-shows and flow of replacement referrals appropriately codified in planning documents?

Practice Task 2 Unmanned study cafe space reservation and IoT control service

📥 Assignment input (Input)

  • • study-cafe-meeting-notes.md meeting minutes
  • • Codex writing prompts learned in previous exercises in steps 1-5.

📤 Assignment output

  • • study-cafe-requirements.md (IoT-linked exception handling integrated requirements definition)
  • • Hardware-Software Control Verification Scenario Specification

💡 Mission background and task goals

These are the meeting minutes of a service that allows study room reservations and remote control of offline hardware (door locks, smart power cuts, thermostats, etc.) through mobile devices in an unmanned study cafe. A complete IoT business specification must be established by deriving multiple safety processes, such as dynamic creation of a pin code at the time of payment completion, synchronization of IoT device status and pin code operation restriction conditions upon entry/exit, and emergency exception door opening and alarm notification in the event of an IoT communication failure (door lock not working, etc.).

🛠️ Self-practice guide

  1. Download meeting minutes: Click the button below study-cafe-meeting-notes.md Download the original file.
  2. Requirements Documentation: Based on the downloaded meeting minutes, instruct Codex AI to specify user and hardware flows (activation 10 minutes before the reservation time, notification 5 minutes before end, power off upon end).
  3. Refine the fault exception process: Exception response measures for each scenario that may occur, such as IoT communication interruption or door lock malfunction (opening of the server's spare network, emergency notification of manager transmission failure, emergency emergency call, etc.) are clearly written by applying multiple fail-safe design elements.

Learner self-practice materials

Click the button below to download the original meeting minutes to complete Mission 2.

Recommended AI Practice Prompts

Mission 2 Prompt[Role: IoT Systems Business Analyst]
Based on the study-cafe-meeting-notes.md file you downloaded, fill out study-cafe-requirements.md including the following conditions.

1. Define the normal flow of study room reservation and payment
2. Hardware control timing specification (password activation 10 minutes before reservation time and air conditioner turned on, audio announcement 5 minutes before end, immediate shutdown upon end)
3. Exception and fail-safe flow:
   - Automatic switching to spare wireless network and emergency remote opening process in case of offline IoT door lock network failure
   - Emergency manual door opening button provided when pin code transmission fails and immediate dashboard alarm creation rule for administrators
4. Requirements verification criteria

Organize the flow very specifically in the form of a timeline so that developers and HW engineers can work together without ambiguity.

🎯 Key evaluation and verification criteria

• Are the timeline conditions for hardware (door locks, lighting, heating and cooling) control specified before/after the reservation time?
• Are dual emergency response procedures in the event of a physical failure, such as IoT device network failure, included in the requirements definition?
• Are reservation business rules, such as non-member reservation restrictions and extended payment policies, expressed in formal sentences?