mirror of
https://github.com/coleam00/context-engineering-intro.git
synced 2025-12-17 17:55:29 +00:00
45 lines
2.5 KiB
Markdown
45 lines
2.5 KiB
Markdown
# Product Requirement Prompt (PRP) Concept
|
|
|
|
"Over-specifying what to build while under-specifying the context, and how to build it, is why so many AI-driven coding attempts stall at 80%. A Product Requirement Prompt (PRP) fixes that by fusing the disciplined scope of a classic Product Requirements Document (PRD) with the “context-is-king” mindset of modern prompt engineering."
|
|
|
|
## What is a PRP?
|
|
|
|
Product Requirement Prompt (PRP)
|
|
A PRP is a structured prompt that supplies an AI coding agent with everything it needs to deliver a vertical slice of working software—no more, no less.
|
|
|
|
### How it differs from a PRD
|
|
|
|
A traditional PRD clarifies what the product must do and why customers need it, but deliberately avoids how it will be built.
|
|
|
|
A PRP keeps the goal and justification sections of a PRD yet adds three AI-critical layers:
|
|
|
|
### Context
|
|
|
|
- Precise file paths and content, library versions and library context, code snippets examples. LLMs generate higher-quality code when given direct, in-prompt references instead of broad descriptions. Usage of a ai_docs/ directory to pipe in library and other docs.
|
|
|
|
### Implementation Details and Strategy
|
|
|
|
- In contrast of a traditional PRD, a PRP explicitly states how the product will be built. This includes the use of API endpoints, test runners, or agent patterns (ReAct, Plan-and-Execute) to use. Usage of typehints, dependencies, architectural patterns and other tools to ensure the code is built correctly.
|
|
|
|
### Validation Gates
|
|
|
|
- Deterministic checks such as pytest, ruff, or static type passes “Shift-left” quality controls catch defects early and are cheaper than late re-work.
|
|
Example: Each new funtion should be individaully tested, Validation gate = all tests pass.
|
|
|
|
### PRP Layer Why It Exists
|
|
|
|
- The PRP folder is used to prepare and pipe PRPs to the agentic coder.
|
|
|
|
## Why context is non-negotiable
|
|
|
|
Large-language-model outputs are bounded by their context window; irrelevant or missing context literally squeezes out useful tokens
|
|
|
|
The industry mantra “Garbage In → Garbage Out” applies doubly to prompt engineering and especially in agentic engineering: sloppy input yields brittle code
|
|
|
|
## In short
|
|
|
|
A PRP is PRD + curated codebase intelligence + agent/runbook—the minimum viable packet an AI needs to plausibly ship production-ready code on the first pass.
|
|
|
|
The PRP can be small and focusing on a single task or large and covering multiple tasks.
|
|
The true power of PRP is in the ability to chain tasks together in a PRP to build, self-validate and ship complex features.
|