Codex for Student Projects

Activity 1: First Codex Action

Task

Ask Codex to make a simple HTML file and view it in the preview.

Instructions

1. Open the folder.
2. Switch the model and thinking level as instructed in the session.
3. Type or copy the prompt.

Make a simple HTML page that says "Hello world" in big sparkly letters.

Reflection

• Reflect on the design choices Codex made.
• Notice what Codex added or left out.
• Try a different model or instruction and compare the result.

Activity 2: Organise a Folder

Task

Organise a folder of messy files.

Instructions

1. Open messy-research-archive/.
2. Ask Codex what is in the folder.
3. Ask Codex to suggest an organisation.
4. Ask Codex to make a visualisation.

Activity 3: Catalog a Folder of Images

Task

Create a catalog of a folder with image files.

Instructions

1. Open the folder as a Codex project.
2. Ask what is in the folder.
3. Ask Codex to rename the images in order.
4. Ask Codex to build a single HTML catalog with images.

Activity 4: Explore AGENTS.md

Task

Explore a fuller, generic AGENTS.md sample.

Sample AGENTS.md

# Agent Instructions: Local Project Workspace

## Operating Mode

- Folder type: local project workspace.
- Audience: project owner and collaborators.
- Agent role: inspect, plan, edit, explain.
- Default action: inspect first; ask before broad changes.

## First Moves

- Read `README.md`, `AGENTS.md`, and visible task files.
- List relevant files before editing.
- Confirm target files and intended changes.

## Behaviour

- Change size: small, reversible.
- Preserve user work.
- Rename/delete/move many files: ask first.
- Destructive or bulk commands: explain first.
- Secrets: never write into Markdown, logs, screenshots, or project files.

## Logging

- If `_TASK_LOG/actions.jsonl` exists: append after meaningful changes.
- If no log exists: ask before creating one.
- Format: JSONL, one event per line.
- Keep entries short.

Example:

```json
{"timestamp":"2026-06-29T09:00:00+01:00","action":"Updated project notes","files_changed":["notes/project-summary.md"],"checks":["reviewed generated HTML"]}
```

## Checks

- Run relevant local checks when available.
- State what was checked.
- State what was not checked.

## Handoff

- Summarise changes.
- List open questions.
- Suggest next steps only when useful.

## Public-Safe Boundary

- No private paths.
- No real participant data.
- No access tokens.
- No internal screenshots.
- No personal notes unless user confirms private use.

Instructions

1. Open sample-project/.
2. Read AGENTS.md.
3. Notice the operating mode, behaviour rules, logging format, reflection rule, and public-safe boundary.
4. Ask Codex what it would do differently because of this file.

Activity 5: Explore Skills

Task

Explore how skills extend what an agent can do.

Links

• AgentSkills.io
• Anthropic public skills
• Matt Pocock's skills
• Dominik's agent skills
• Matt Pocock's TEACH skill

Local Skill Copies

• Skills folder README
• teach/SKILL.md
• grill-with-docs/SKILL.md
• handoff/SKILL.md
• to-prd/SKILL.md
• dominiks-agent-skills/academic-pdf-to-mkd/SKILL.md
• dominiks-agent-skills/agents-md-streamline/SKILL.md
• dominiks-agent-skills/claim-fact-checker/SKILL.md
• dominiks-agent-skills/codex-images-workflow/SKILL.md
• dominiks-agent-skills/cognitive-walkthrough/SKILL.md
• dominiks-agent-skills/paper-reviewer-skill/SKILL.md
• dominiks-agent-skills/project-changelog/SKILL.md
• dominiks-agent-skills/project-setup/SKILL.md
• dominiks-agent-skills/single-html/single-html-document/SKILL.md

What To Notice

• AGENTS.md is placed in a project folder.
• SKILL.md is placed in its own folder.
• Agents load AGENTS.md automatically when working in the folder.
• Agents load SKILL.md on request or when they think they need it.

Sample SKILL.md

---
name: teach
description: Teach the user a new skill or concept, within this workspace.
disable-model-invocation: true
argument-hint: "What would you like to learn about?"
---

The user has asked you to teach them something. This is a stateful request - they intend to learn the topic over multiple sessions.

## Teaching Workspace

Treat the current directory as a teaching workspace. The state of their learning is captured in this directory in several files:

- `MISSION.md`: A document capturing the _reason_ the user is interested in the topic. This should be used to ground all teaching.
- `./reference/*.html`: A directory of reference materials. These are the compressed learnings from the lessons.
- `RESOURCES.md`: A list of resources which can be explored to ground your teaching in contextual knowledge.
- `./learning-records/*.md`: A directory of learning records, which capture what the user has learned.
- `./lessons/*.html`: A directory of lessons. A lesson is a single, self-contained HTML output that teaches one tightly-scoped thing tied to the mission.
- `NOTES.md`: A scratchpad for user preferences or working notes.

## Philosophy

To learn at a deep level, the user needs three things:

- Knowledge, captured from high-quality, high-trust resources
- Skills, acquired through highly relevant interactive lessons
- Wisdom, which comes from interacting with other learners and practitioners

Before `RESOURCES.md` is well-populated, find high-quality resources which will help the user acquire knowledge. Never trust your parametric knowledge.

## Lessons

A lesson is the main thing you produce. Each lesson is one self-contained HTML file, saved to `./lessons/` and titled `0001-dash-case-name.html` where the number increments each time.

The lesson should be short, useful, and tied to the mission. Each lesson should give the user a single tangible win that they can build on.

## The Mission

Every lesson should be tied into the mission: the reason that the user is interested in learning about the topic.

If the user is unclear about the mission, or `MISSION.md` is not populated, first ask why they want to learn this.

## Zone Of Proximal Development

Each lesson should challenge the user just enough.

If the user does not specify exactly what they want to learn, read their learning records and choose the next useful lesson based on their mission.

## Knowledge

Lessons should be designed around a skill the user is going to learn. The knowledge in the lesson should be only what is required to acquire that skill.

Knowledge should first be gathered from trusted resources. Use `RESOURCES.md` to keep track of them.

## Skills

Skills should be taught through interactive lessons. Use retrieval practice, light in-browser tasks, or guided real-world steps.

Each lesson should use a feedback loop where the user receives feedback on their performance.

## Reference Documents

While creating lessons, also create reference documents. Lessons can reference these documents, and the user can return to them later.

## NOTES.md

Record user preferences in `NOTES.md` so they can shape future lessons.

Activity 6: Biodiversity Data Feasibility

Task

Assess whether AI can help improve biodiversity in a local area based on data. Use Stonesfield in Oxfordshire as the feasibility test.

Project Description

• Aim: Assess whether AI can help improve biodiversity in a local area based on a dataset.
• Test area: Stonesfield, Oxfordshire.
• Question: What public data can be used now, what first result can it produce, and what would still be needed before suggesting practical improvements?
• Desired output: A first evidence-based view of what has already been observed, where the gaps are, and what further work could help.

Workflow

1. Find useful data.
2. Download a sample.
3. Visualise the sample.
4. Decide what else is possible.
5. Explain the process.
6. Set up project memory.

Learning Log

Open `_LEARNING_LOG` as a separate project in Codex. Start a new chat and use voice or typing to record reflections and questions for later.

Export Notes Clear Notes