Why this page exists
Session 5 was the final integration session. Instead of introducing one more tool, the focus was to combine everything from the series into one repeatable workflow for coursework-level Python projects.
This page turns that final workflow into a checklist you can reuse.
The Session 5 workflow in one line
Read first, structure second, automate third, verify always.
Step-by-step playbook
1. Set up the project folder
- Create one dedicated project directory.
- Put all downloaded coursework files in that directory.
- Keep raw data separate from derived outputs.
2. Read documents before coding
- Read the coursework brief fully.
- Read the data README for variable descriptions.
- Confirm deliverables and submission format before implementation.
3. Create a short AGENTS.md
- Define your project goal and scope.
- Add coding and data rules.
- Keep it compact and actionable.
- Include a non-negotiable rule such as: never overwrite raw CSV files.
If you need a template, use the AGENTS.md Writing Guide.
4. Define high-value skills
Start with one repeated workflow, such as data cleaning.
A practical skill pattern:
- inspect data quality
- generate a report
- propose fixes
- ask for approval
- apply only approved fixes
5. Execute in small chunks
- Do not ask for the full project in one prompt.
- Build incrementally and review after each step.
- Keep commits or checkpoints frequent so you can recover quickly.
6. Verify before accepting outputs
- Check code correctness and reproducibility.
- Check economic plausibility of results.
- Check whether written answers match the actual outputs.
Skills vs sub-agents: when to use each
Common cases:
- Repeated, structured task (cleaning, formatting, diagnostics): use a skill for predictable workflow and easy reuse.
- Focused specialist task with autonomy (consistency review, cross-file checks): use a sub-agent for dedicated reasoning.
- Large project with independent streams: use multiple agents carefully, with clear coordination rules.
Practical recommendation from Session 5:
- start with skills
- add one sub-agent for consistency checks
- only scale to many agents if your structure is already clear
Final session reminders
- You are responsible for the final output, even if AI generated part of it.
- Do not submit content you have not read and verified.
- Keep your economics reasoning in the driver's seat.