source-driven-development
by addyosmaniThe source-driven-development skill grounds framework-specific coding in official docs, helping you verify patterns before you implement. It is ideal for source-driven-development usage in React, Vue, Next.js, Svelte, Angular, and similar stacks when correctness, provenance, and version-sensitive decisions matter.
This skill scores 82/100, which means it is a solid listing candidate for directory users who want source-cited, documentation-grounded implementation guidance. The repository gives enough operational clarity to justify installation for framework-specific coding where correctness matters, though users should expect a doc-first workflow rather than an automated or tool-backed skill.
- Clear triggerability: the description explicitly says to use it for framework/library work where authoritative, source-cited code matters.
- Strong operational guidance: the SKILL.md lays out a stepwise DETECT → FETCH → IMPLEMENT → CITE process and defines when not to use it.
- Good install decision value: the repo is substantial (7.8k body, multiple headings, no placeholder markers) and focuses on reducing outdated or memory-based implementation errors.
- No support files or scripts are included, so execution depends entirely on the written instructions rather than a bundled workflow.
- It is broad and documentation-centric, so users looking for speed or general logic tasks may find it unnecessary or overkill.
Overview of source-driven-development skill
What source-driven-development does
The source-driven-development skill helps you write framework-specific code only after checking the official docs, then cite or mirror those sources in the implementation. It is for work where “probably correct” is not good enough: starter code, recurring app patterns, upgrades, and any feature whose right answer changes by framework version.
Who should use it
Use the source-driven-development skill if you are building with React, Vue, Next.js, Svelte, Angular, or similar tools and want fewer stale patterns. It is especially useful for agents generating code for forms, routing, auth, state, data fetching, and project scaffolding.
Why it is different from a normal prompt
A generic prompt may produce plausible code from training data. This skill pushes the workflow toward verify-first behavior: detect the stack, check the official source, then implement from that source. That makes it better for production work, version-sensitive APIs, and code review situations where provenance matters.
How to Use source-driven-development skill
Install the skill in your agent setup
Use the repository’s install flow for skills, then load source-driven-development into the agent context before asking for code. A typical source-driven-development install looks like the repo’s skill manager command, followed by reading the skill file itself so the agent can follow the intended workflow.
Turn your task into a source-first prompt
For best source-driven-development usage, give the model the framework, version if known, runtime, and exact goal. Strong input looks like: “Use source-driven-development to add a protected dashboard route in Next.js App Router 15, using official docs only, and explain the source for each pattern.” Weak input is just “build auth.” The more specific the stack and feature, the less the agent has to guess.
Read the right files first
Start with SKILL.md, then inspect any linked docs, examples, or repository notes the skill references. For this repository, there are no helper folders to chase, so the main job is to follow the process in SKILL.md closely and map it to the target app’s own conventions.
Workflow that improves output quality
Use a 4-step loop: identify the stack and version, fetch the official documentation, implement the smallest correct version, then verify that the result still matches the docs. If you are asking for code generation, include constraints such as existing folder structure, package manager, framework mode, and whether you need citations in comments or in a separate explanation.
source-driven-development skill FAQ
Is source-driven-development only for large frameworks?
No. The source-driven-development guide is most valuable when APIs change often or have multiple competing patterns, which is common in modern frameworks and libraries. It is less useful for plain language changes or logic that does not depend on framework version.
Should I use it instead of a normal code prompt?
Use it when correctness and traceability matter more than speed. Ordinary prompts are fine for simple, version-agnostic edits, but the source-driven-development for Code Generation approach is better when you want code that can be defended against current official docs.
Is it beginner-friendly?
Yes, if you can name the framework and describe the feature. Beginners benefit because the skill reduces pattern guesswork, but they still need to provide the app stack and target behavior. Without that, the agent may verify the wrong docs.
When is it a bad fit?
Do not use it for trivial edits, pure logic, or situations where the user explicitly wants the fastest possible answer without verification. If the task does not depend on a specific framework version, the extra source-checking usually adds overhead without much value.
How to Improve source-driven-development skill
Give the agent better source targets
The biggest quality gain comes from naming the exact framework, version, and feature. For example, say “official Next.js App Router docs for server actions” instead of “Next.js stuff.” Better targets reduce wrong-source lookups and improve the final code’s accuracy.
State the output constraints up front
If you need citations, explain the format: inline comments, a short source note, or a brief implementation summary. Also specify what must not change, such as existing types, file names, or UI behavior. That helps the skill stay source-driven without overrewriting your codebase.
Watch for common failure modes
The main risks are using the wrong version of a doc, mixing patterns from multiple releases, or overfitting the implementation to a tutorial example. If the first answer feels generic, ask the agent to re-check the official docs and narrow the solution to your exact stack.
Iterate from source, not from guesswork
If the first result is close but not ideal, ask for a second pass that preserves the verified parts and only adjusts the missing behavior. For source-driven-development, the best improvement loop is: clarify the stack, confirm the source, then refine the implementation with one specific change at a time.