source-driven-development

Source-Driven Development

Overview

Read authoritative sources before writing code that depends on them. AI training data has a cutoff — and even before that cutoff, documentation often reflects correct usage better than the patterns in training data. When building against a library, framework, or API, consulting the real source prevents hallucinated APIs, outdated patterns, and subtle bugs from subtle version differences.

When to Use

• Using a library or framework that may have changed since training
• Unsure about current API signatures, hooks, or configuration
• Working with rapidly evolving tools (AI SDKs, cloud APIs, build tools)
• Implementing against a third-party API (authentication flows, endpoints, payloads)
• Encountering an error that suggests an API mismatch

When NOT to use:

• Writing pure algorithms or logic with no external dependencies
• Working with extremely stable, well-established APIs (e.g., basic array methods)
• You have high confidence the pattern is current and correct

The Source-Driven Protocol

Step 1: Identify What Needs Verification

Before writing code that uses an external library or API, list the assumptions you're making:

ASSUMPTIONS TO VERIFY:
- Does this hook/function/method still exist with this name?
- Are the arguments in the right order?
- Is this the recommended pattern (not deprecated)?
- Are there new required parameters or options?
- Does the return type match what I expect?

Step 2: Fetch Authoritative Sources

Use Agent Zero tools to retrieve current documentation:

For package documentation (npm):

document_query: https://www.npmjs.com/package/<package-name>

For GitHub repos:

browser: open → navigate to github.com/<owner>/<repo>/blob/main/README.md
browser: content → read the README

For official docs sites:

browser: open → navigate to the official docs URL
browser: content → read the relevant section

For specific API references:

document_query: <direct URL to the API reference page>