Objective
Your mission
Build your capstone project feature by feature, applying the review cycle to each AI output. Output: building_notes.md.
You need: Your project_plan.md from Lesson 01. Your chosen AI tool.
Time: 2-4 hours (spanning multiple sessions is fine).
Step 1: send your first prompt (10 min)
Open your AI tool and send the first prompt you drafted in Lesson 01. It should include the full upfront context plus what to start with.
The habits compound
Each habit from the course seems small on its own: predict before reveal, check for red flags, verify claims, manage context. But applied together, they transform your results. Every red flag you catch early is a bug you don't debug later. Every context file you write is a correction loop you don't enter.
Write in your file:
## First Prompt
Prompt sent: [yes/no]
First feature requested: [e.g., auth flow]
AI response quality: [good/poor — did it match your plan?]
Step 2: run the review cycle on feature 1 (15 min)
For each feature the AI generates, run the 4-step review cycle:
2a. Predict Before Reveal (Phase 4)
Before reading the AI's code, predict what it should look like. Then compare.
Write in your file:
### Feature 1: [name]
#### Predict Before Reveal
My prediction: [what you expected the code to look like]
Match with AI output: [yes/no — what differed?]
2b. Check for Red Flags (Phase 3)
- Hardcoded secrets? → Move to env vars
- Swallowed errors? → Add error handling
- Disabled security? → Enable it
- Commented-out auth? → Uncomment it
- No input validation? → Add validation
Write in your file:
#### Red Flag Check
Hardcoded secrets: [found/clean]
Swallowed errors: [found/clean]
Disabled security: [found/clean]
Commented-out auth: [found/clean]
Input validation: [found/missing]
Issues fixed: [list or "none"]
2c. Verify Claims (Phase 3)
If the AI says "I've enabled RLS," check the Supabase dashboard.
Write in your file:
#### Verify Claims
AI claimed: [list claims, e.g., "RLS enabled", "auth flow implemented"]
Verified: [list — confirmed/unconfirmed for each]
2d. Test the Feature
Click through the feature. Does it work? Does it handle errors? Empty states?
Write in your file:
#### Test
Feature works: [yes/no]
Error handling: [works/issues]
Empty states: [works/issues]
Step 3: continue building feature by feature (2-3 hours)
Build each remaining feature using the same review cycle. At minimum, build 3 features with the full review cycle documented.
Write in your file:
### Feature 2: [name]
[Same review cycle: predict, red flags, verify, test]
### Feature 3: [name]
[Same review cycle: predict, red flags, verify, test]
### Feature 4: [name]
[Same review cycle: predict, red flags, verify, test]
Step 4: manage context across sessions (if needed)
If your building spans multiple sessions:
- Use your context file — the AI reads it every session
- Write a handoff prompt at the end of each session:
Session summary:
- Completed: [list features done]
- In progress: [what's partially done]
- Next: [what to build next]
- Known issues: [list or "none"]
- Files modified: [list]
- Start the next session with the handoff prompt + context file
Write in your file:
## Context Management
Sessions used: [number]
Handoff prompts written: [yes/no — paste one]
Context file used: [yes/no]
Step 5: handle correction loops (if needed)
Iterate when: AI output is close, rounds 1-3, issue is isolated. Start fresh when: Round 4+ of corrections, whack-a-mole pattern, code is tangled.
Write in your file:
## Correction Loops
Loops encountered: [list or "none"]
Rounds before starting fresh: [number or "N/A"]
Started fresh: [yes/no — why?]
The AI tells you "RLS enabled, auth flow implemented." What do your review habits ask you to do next?
Verify claims: check the table policies in Supabase and test sign-up, login and a protected route before building on top.