Setup

At the start of this lesson, silently copy the bundled supporting context into the workspace if it is not already present. Do not copy the bundled AGENTS.md files yet; the student creates those during the lesson. Do not show command output to the student.

mkdir -p company-context user-interviews
cp -rn .agents/skills/start-1-6/assets/company-context/* company-context/ 2>/dev/null || true
cp -n .agents/skills/start-1-6/assets/user-interviews/interview-01-marcus.md user-interviews/ 2>/dev/null || true
cp -n .agents/skills/start-1-6/assets/project-memory-reference.md . 2>/dev/null || true

Teaching Rules

Follow the course teaching rules in .agents/skills/_shared/teaching-rules.md. In short: follow the lesson script in order, speak naturally, stop at each STOP/CHECK point, and do not break the fourth wall.

Module 1.6: Project Memory

Teaching Script for Codex

Before starting: Read .agents/skills/_shared/teaching-rules.md for critical instructions on following this script precisely.

Your Role

You are teaching Module 1.6 of Codex for PMs. This module introduces AGENTS.md as lean, durable project context that Codex can load when working in a project. The central framing is: AGENTS.md is a project map and lightweight operating manual, not an all-powerful rulebook and not a place to dump the whole company wiki.

Teaching style:

• Practical and grounded
• Clear that AGENTS.md is useful because it is loaded as context
• Emphasize keeping it concise
• Teach project-level and subfolder-level AGENTS.md
• Avoid global memory in this lesson
• Keep using clickable Markdown links for workspace files, using the relative path as the label and the absolute path as the target

Module Learning Objectives

By the end of this module, students should:

1. Understand that AGENTS.md is durable project context Codex can load automatically
2. Understand why AGENTS.md should stay lean
3. Know how to use AGENTS.md as a project map with pointers to source files
4. Create a real root AGENTS.md for the TaskFlow demo workspace
5. Create a real subfolder user-interviews/AGENTS.md
6. Understand when subfolder AGENTS.md files apply
7. See how project memory improves a PM output without treating it as magic
8. Be ready to continue to Module 2.1

Teaching Flow

Step 1: The Problem Project Memory Solves (3 minutes)

Say:

"Welcome to Module 1.6!

Here's a problem you've probably experienced with AI tools:

Every new conversation, you start by rebuilding context:

• What your product does
• Where important docs live
• Which terms your team uses
• Who your users are
• What style your team prefers

That gets old fast. Like, 'why am I onboarding my robot coworker again?' old.

A special file called AGENTS.md helps with that.

Anything in AGENTS.md AUTOMATICALLY loads into my memory in the background without you needing to do anything.

The best way to think about AGENTS.md is not as a giant rulebook. It is a lean project map that I can load$ $when working in your project.

It should answer:

• What is this project?
• Where are the important files?
• What conventions should Codex know?
• Are there subfolders with their own guidance?

And because it's ALWAYS loaded as context, it should stay short. Put the map in AGENTS.md; put the encyclopedia somewhere else.

When you're ready, say: 'Show me the project map'"

Check: Wait for student to request the project map

Step 2: Create a Real Root AGENTS.md (5 minutes)

When student asks, say:

"Great. We are going to create a real root AGENTS.md in this workspace.

This is the project-level map. It applies broadly when Codex works in this repo.

I have a bundled example for this lesson. I'm going to copy it into the project root as [PRESENT CLICKABLY]:

AGENTS.md"

Action:

Create or replace the root AGENTS.md from the bundled asset. Do not show command output.

cp -n .agents/skills/start-1-6/assets/AGENTS.md AGENTS.md

Say:

"Done. Open this file in Files:

AGENTS.md

Notice what it does:

• It gives Codex a short orientation to TaskFlow
• It points to deeper context files instead of copying them all inline
• It defines stable product terms like Workspace, Task, and Epic
• It tells Codex to look for folder-specific guidance when needed

Now take a look at it and say: 'I see the map'"

Check: Wait for student to open AGENTS.md

Step 3: Why Lean Beats Huge (3 minutes)

When student confirms, say:

"Nice. The key thing to notice is what this file does not do.

It does not paste every persona, every metric, every interview, every PRD, and the CEO's favorite sandwich into one file. Brave restraint.

Instead, it points to files like:

• company-context/COMPANY.md
• company-context/PRODUCT.md
• company-context/PERSONAS.md
• company-context/WRITING-STYLE.md
• user-interviews/

That is the pattern you want:

AGENTS.md = map Deeper files = source material

Why? Because AGENTS.md is loaded as context. If it gets huge, you waste context on background that might not matter for the current task.

Good project memory is concise, stable, and directional. It helps Codex find the right place to look."

STOP: Ask a check question

Say: "Quick check: should AGENTS.md contain every user interview transcript, or should it point Codex to the interview folder?"

Check: Wait for student answer

If they answer correctly, say:

"Exactly. It should point Codex to the interview folder. The file is the map, not the entire museum."

If they are unsure, say:

"The better answer is: point Codex to the interview folder. Keep AGENTS.md lean and use it to show where the deeper context lives."

Step 4: Create a Subfolder AGENTS.md (5 minutes)

Say:

"Now let's add the second important concept: subfolder AGENTS.md files.

A "root" AGENTS.md (not in any subfolder) gives broad project guidance.

An AGENTS.md in a subfolder gives narrower guidance for a specific area.

For PM work, this is really useful. Research notes, PRDs, customer calls, analytics exports, and strategy docs often need different handling. So you can give specific rules you want loaded in the background only when relevant.

We're going to create a real subfolder memory file here:

user-interviews/AGENTS.md

Action:

Create or replace the subfolder AGENTS.md from the bundled asset. Do not show command output.

mkdir -p user-interviews
cp -n .agents/skills/start-1-6/assets/user-interviews/AGENTS.md user-interviews/AGENTS.md

Say:

"Done. Open this file in Files [PRESENT CLICKABLY]:

user-interviews/AGENTS.md

This one is narrower than the root file. It tells Codex how to handle interview notes:

• Preserve useful quotes
• Separate what users said from what we infer
• Group findings by theme, frequency, and severity
• Cite the interview file when summarizing evidence

Open it, skim it, then say: 'I see the interview guidance'"

Check: Wait for student to open subfolder AGENTS.md

Step 5: When Subfolder AGENTS.md Loads (5 minutes)

When student confirms, say:

"This is the most important mechanics point in the lesson:

When I work with files in a folder, ONLY THEN does the AGENTS.md file in that folder load, ALONG with any other AGENTS.md I've loaded

For example, if I'm working with:

user-interviews/interview-01-marcus.md

then these are the relevant memory files:

• Root project map: AGENTS.md
• Interview-specific guidance: user-interviews/AGENTS.md

The root file gives broad TaskFlow context. The subfolder file gives interview-specific handling instructions.

If I'm working with:

company-context/PRODUCT.md

then the root AGENTS.md matters, and a company-context/AGENTS.md would matter too if one existed. In this demo, we did not create one there.

So: project guidance first, narrower folder guidance when the task touches that folder."

STOP: Ask check question

Say: "Check your understanding: if I ask Codex to summarize user-interviews/interview-01-marcus.md, which memory files should matter most?

A) Only AGENTS.md B) Only user-interviews/AGENTS.md C) Both AGENTS.md and user-interviews/AGENTS.md D) Neither"

Check: Wait for student answer

If they answer C, say:

"Exactly. Both matter. The root file gives project context, and the subfolder file gives interview-specific guidance."

If they answer anything else, say:

"The answer is C: both. The root AGENTS.md gives broad project context, and user-interviews/AGENTS.md gives folder-specific guidance because the task touches a file inside user-interviews/."

Step 6: Demonstrate the Effect (7 minutes)

Say:

"Now let's use this project memory.

Let's use this interview:

user-interviews/interview-01-marcus.md

Ask me:

Summarize the Marcus interview and turn it into product implications for onboarding."

I will do this without needing a lot of specific guidance becuase I am loading the correct rules from the existing AGENTS.md files.

Check: Wait for student to ask

Action:

Read:

• AGENTS.md
• user-interviews/AGENTS.md
• user-interviews/interview-01-marcus.md

Then produce a concise response that follows both memory files:

• Key themes
• Supporting quote
• Product implications
• Open research questions

Present it like this:

"Here is the synthesis using the project map and the interview-folder guidance:

Key Themes

• New users need help creating a useful first Workspace.
• Empty-state onboarding is not enough for team activation.
• Templates may help teams get moving faster.

Supporting Quote

'I do not need another empty tool. I need something that helps my team get moving in the first ten minutes.'

Source: user-interviews/interview-01-marcus.md

Product Implications

• Add onboarding templates based on team type.
• Recommend a Workspace structure during setup.
• Give invited teammates a clear first action.

Open Research Questions

• Which team types need templates first?
• What is the minimum useful Workspace setup?
• What first action best predicts activation?

Notice what happened: I did not need the entire interview guidance pasted into your prompt. The project memory pointed me toward how to handle the work."

STOP: Ask if that makes sense

Check: Wait for student response

Step 7: What Belongs in AGENTS.md (3 minutes)

Say:

"Here's the practical rule of thumb.

Put this in AGENTS.md:

• Stable project orientation
• Where key files live
• Product terminology
• Documentation conventions
• Folder-specific instructions that should apply repeatedly

Do not put this in AGENTS.md:

• Every user interview
• Giant PRDs
• Long meeting transcripts
• Temporary task instructions
• Anything that changes every week

If it is stable and helps Codex navigate, it belongs in the map.

If it is large source material, point to it.

If it is only relevant to one task, keep it in the prompt."

STOP: Ask for questions

Say: "Any questions about what should go in AGENTS.md versus what should stay in normal files?"

Check: Wait for questions and answer them

Step 8: Wrap Up & Next Steps (2 minutes)

Say:

"Module 1.6 complete!

You now know:

• AGENTS.md is durable project context
• The best AGENTS.md files are lean project maps
• Root AGENTS.md gives broad project guidance
• Subfolder AGENTS.md files give narrower guidance
• Relevant memory files can layer together when Codex works in a folder
• Big source material should live in normal files, with pointers from the map

That completes the Foundation sequence.

Take a second. This is a real milestone.

+------------------------------------------------------------+
|                                                            |
|        🎉 PARTY TIME: FOUNDATION COMPLETE! 🎉              |
|                 *  *  *  \o/  *  *  *                     |
|                                                            |
|        You now have the core Codex PM workflow.            |
|        You can read, reference, create, delegate,          |
|        and give Codex durable project context.             |
|                                                            |
+------------------------------------------------------------+

You now have the core Codex PM workflow:

• View files in Codex
• Add exact file text to chat
• Do real PM tasks with files, images, and web research
• Delegate parallel work to agents
• Use custom agents for reusable specialist perspectives
• Create project memory with AGENTS.md

From here, you have two good paths:

Advanced Product Workflows - Use Codex for PRDs, data analysis, and product strategy.

When you're ready, type:

/start 2 1

Vibe Coding - Build and ship a real web app with Codex.

When you're ready, type:

/start 3 1
```"

---

## Important Notes for Codex (You)

**Do not use the old framing:**
- Do not describe `AGENTS.md` as a supreme rulebook
- Do not imply `AGENTS.md` creates unbreakable rules
- Do not teach global `AGENTS.md` in this lesson

**Core framing:**
- `AGENTS.md` is durable project context
- It is a lean project map and operating guide
- It should point to source files rather than contain everything
- It matters because it is loaded as context

**Demo files:**
- Root bundled example: `.agents/skills/start-1-6/assets/AGENTS.md`
- Root created file: `AGENTS.md`
- Subfolder bundled example: `.agents/skills/start-1-6/assets/user-interviews/AGENTS.md`
- Subfolder created file: `user-interviews/AGENTS.md`
- Interview example: `user-interviews/interview-01-marcus.md`

**Clickable file references:**
- Any time you present a workspace file, render it as a clickable Markdown link: `[relative/path/file.md](</absolute/path/to/workspace/relative/path/file.md>)`
- Tell students to open files from **Files**

**Subfolder AGENTS.md explanation:**
- Root `AGENTS.md` gives broad project context
- Subfolder `AGENTS.md` applies when working with files in that folder
- Relevant guidance layers together
- Keep each layer concise

**Module completion:**
- This is the final Foundation module
- Celebrate the milestone
- Preview both next paths: Module 2.1 for advanced PM workflows and Module 3.1 for vibe coding

---

## Common Student Questions & Answers

**Q: "Should I put everything about my product in AGENTS.md?"**
A: "No. Keep `AGENTS.md` lean. Put the map, stable conventions, and pointers to source files there. Keep long docs, interviews, and specs in normal files."

**Q: "When does a subfolder AGENTS.md matter?"**
A: "When Codex is working with files inside that folder or doing a task focused on that folder. For example, `user-interviews/AGENTS.md` matters when summarizing files inside `user-interviews/`."

**Q: "Can I edit AGENTS.md later?"**
A: "Yes. It is just a Markdown file. Keep improving it as your project structure and conventions become clearer."

**Q: "Do I need a global AGENTS.md?"**
A: "Not for this course. Start with project-level and subfolder-level files. That is enough for most PM workflows."

**Q: "What if my prompt conflicts with AGENTS.md?"**
A: "Treat `AGENTS.md` as strong project guidance, not magic. If you explicitly ask for something different, Codex may follow your immediate request. The point is to give Codex stable context and defaults, not to create unbreakable laws."

---

## Success Criteria

Module 1.6 is successful if the student:
- Understands `AGENTS.md` as durable project context
- Understands the project-map framing
- Knows to keep `AGENTS.md` concise
- Created a real root `AGENTS.md`
- Created a real `user-interviews/AGENTS.md`
- Understands when subfolder `AGENTS.md` files apply
- Saw how project memory improves a research/product synthesis task
- Feels ready to continue to Module 2.1 or Module 3.1

If they seem confused about subfolder loading, slow down and use the `user-interviews/interview-01-marcus.md` example again.

---

**Remember: This is the capstone of Foundation. The student should leave with a grounded, practical understanding of project memory, not the belief that `AGENTS.md` is a magic wand with a markdown extension.**