Talent Prompting with MasterFabric
Structured prompting so everyone gets consistent, high-quality AI assistance.
Talent Prompting with MasterFabric
We use structured prompting so that everyone—trainees, interns, and experienced engineers—can get consistent, high-quality help from AI. MasterFabric’s approach includes reusable patterns, context, and iteration.
Talent Prompting with MasterFabric
We develop talent with clear prompts, expectations, and feedback. MasterFabric practices help people learn and apply AI-assisted workflows effectively.
Prompt structure
A good prompt gives role, task, context, and constraints. Use this as a mental checklist:
[Role] You are a senior engineer familiar with our stack.
[Task] Add a unit test for the function below.
[Context] Code snippet, file path, or link.
[Constraints] Use Jest, follow our naming convention, max 3 test cases.Keeping this structure in mind (even if you don’t write it verbatim) leads to better, more consistent outputs.
Good prompt vs poor prompt
The same task can yield very different results depending on how you ask. Compare:
Input (prompt):
We use React 18 + TypeScript. Our components live in src/components/ and we use Fumadocs UI for docs.
Task: Add a Card that links to the "AI" docs section. Use the existing Card component from Fumadocs, same pattern as the "Interns Hub" card on the docs index. The href should be ./ai and the description should mention agents and automation.
Do not add new dependencies. Keep the description under 80 characters.Why it works: Role/stack is clear, task is specific, context points to existing pattern, constraints (no new deps, length) avoid scope creep.
Input (prompt):
add a card for AIWhy it fails: No stack, no location, no style. The model may guess the wrong framework, wrong component, or wrong route. You get generic or wrong output and waste time fixing it.
Example: Code review request
When you want AI to review or suggest changes, give it enough context:
[Context]
- Repo: Next.js 14 App Router, TypeScript strict.
- This file: app/api/feedback/route.ts (handles POST /api/feedback).
[Task]
Review this handler for security and error handling. Suggest concrete changes.
Do not change the API contract (request/response shape).
[Code]
[paste the snippet or link to the file]Pro tip
Paste the minimal context that affects the answer: the relevant code, the error message, or the doc section. Long, irrelevant context can dilute quality.
Prompt patterns we use
| Pattern | When to use | Example |
|---|---|---|
| Generate | New code, tests, docs | "Generate a Jest test for parseQuery() using our test utils. Input: valid string, empty string, invalid." |
| Refactor | Same behavior, better structure | "Refactor this to use async/await. Keep the same error handling." |
| Explain | Unfamiliar code or concept | "Explain what this hook does and what happens when deps change." |
| Review | Before opening a PR | "Review for security and edge cases. Focus on the validation logic." |
We document patterns that work well in our internal wiki and update them as we learn.
Context and constraints
We specify context (stack, style, file location) and constraints (no new deps, test coverage, length limits) so outputs align with our standards. Good context reduces back-and-forth and improves quality.
Iteration
We treat prompting as a skill: try, review, refine. If the first answer is off, narrow the task or add one constraint and try again. We share examples and anti-patterns internally so the whole team improves.
Why “talent” prompting
We invest in our talent: helping people write better prompts means better AI assistance for everyone and improves quality across the team.