Codex API

Codex generates code that looks correct but doesn't actually work as intended. You spend more time debugging AI output than writing it yourself.

Action plan: Define tests before asking Codex to implement the feature. This creates a validation loop where Codex can iterate until all tests pass:

<code class="language-python"># test_invoice.py - Write this FIRST
def test_calculate_total_with_tax():
items = [{&quot;price&quot;: 100, "qty": 2}, {"price": 50, "qty": 1}]
assert calculate_invoice_total(items, tax_rate=0.1) == 275.0

def test_calculate_total_empty_items():
assert calculate_invoice_total([], tax_rate=0.1) == 0.0</code></pre>

Then prompt Codex: "Implement calculate_invoice_total to pass all tests in test_invoice.py. Keep iterating until all tests pass."

When AI is involved, tests act as guardrails constraining AI behavior, not just catching bugs. Aim for 100% test coverage of core code.

Every time you start a new Codex session, you have to re-explain your project structure, build commands, and conventions.

Recommended approach: Create an AGENTS.md file in your repo root. Codex loads it into context automatically on every session:

<code class="language-markdown"># AGENTS.md

## Project
Task management API built with Express + TypeScript + Prisma.

## Commands
- - Start dev server
- - Run Jest tests
- - ESLint check

## Conventions
- All API routes in src/routes/, one file per resource
- Use Zod for request validation
- Return consistent error format: { error: string, code: number }
- Never modify migration files directly

## Done when
- All existing tests pass
- New tests written for new functionality
- No lint errors</code></pre>

Pro tip: When Codex makes the same mistake twice, ask it for a retrospective and update AGENTS.md with the fix. Keep the file short — a concise AGENTS.md outperforms a verbose one because it stays within the effective context window.

Issue:
Codex API onboarding fails most often from key mismanagement, not from API availability.

What to do:
1. Create a dedicated API key for each integration environment.
2. Store keys in secret manager, never in shared plaintext notes.
3. Test with one small request before enabling production traffic.

Result:
Auth errors are easier to isolate and deployment reliability improves.

Symptom:
Codex API requests return auth errors even though the key looks valid.

Fix path:
1. Confirm the correct key is used in the correct environment.
2. Rotate the key and remove stale copies from local configs.
3. Send one minimal test request before full workflow runs.

Expected outcome:
Auth failures are isolated quickly and integration returns to normal.