Prompting technique

LLM-Oriented Context Structuring

ReadMe.LLM: A Framework to Help LLMs Understand Your Library, arXiv:2504.09798 (2025) · read the paper

What it does

A method for creating machine-readable documentation for LLMs in the form of a structured file with XML markup, containing usage rules, code examples, and library function descriptions. Submitted to the prompt as context, increasing code generation accuracy from 20-30% to 100% through few-shot learning and explicit instructions.

Submitted to the prompt as context, increasing code generation accuracy from 20-30% to 100% through few-shot learning and explicit instructions.checked against the paper
Task typegeneration
Shapesingle shot
Efforttwo or three steps
Addressesinvented facts and citations; answers that are close but wrong; the model losing what was agreed earlier
Published2025-04-14
AuthorsSandya Wijaya, Jacob Bolano, Alejandro Gomez Soteres et al.

When it helps

Use ReadMeLLM when working with lesser-known libraries, internal tools, or custom APIs that LLMs lack sufficient knowledge about. Especially effective for code generation when standard documentation doesn't yield needed results. Apply to create structured context with rules, examples, and function descriptions. Ideal for integration into RAG systems, CI/CD pipelines, and IDE plugins where high-accuracy code generation without hallucinations is required.

A prompt you can paste

Generic by design: it applies the technique without knowing your task. Adapt the marked parts.

Prompt
# LLM-Oriented Context Structuring (ReadMeLLM) Prompt

# CONTEXT AND RULES FOR LIBRARY UNDERSTANDING

## Library Name: [Placeholder: Name of the library]
## Version: [Placeholder: Version of the library]

## 1. Core Purpose:
Describe the main goal and functionality of the library in 1-2 sentences.
- [Placeholder: Core purpose description]

## 2. Key Components/Modules:
List the main modules or classes and their primary function.
- [Placeholder: Module 1 Name]: [Brief description]
- [Placeholder: Module 2 Name]: [Brief description]
- ...

## 3. Rules for Usage:
Specify constraints and best practices for using the library.
- Always use functions/methods from the provided context.
- If a function is not described, do not invent its behavior.
- Prioritize [Placeholder: Specific function/method] for [Specific task].
- Avoid [Placeholder: Specific anti-pattern or deprecated usage].
- [Add any other crucial rules]

## 4. Core Function Examples:
Provide 1-3 clear, concise examples demonstrating common use cases. Each example should include:
- A brief description of the task.
- The code snippet.
- The expected output or outcome.

### Example 1: [Task Description]

If this one does not fit, the two closest alternatives in the corpus are Debugging Prompts Framework and Context Injection, which target the same failure from a different angle.

Worked example

The same technique applied to a concrete job: review a 2,000-line pull request and report only real defects. Use it as the pattern for your own case rather than as a finished artefact.

Worked example
# LLM-Oriented Context Structuring for Code Review

## 🎯 Goal
To ensure the LLM identifies only genuine defects in a 2,000-line pull request by providing it with a structured, machine-readable context.

## 📖 Context and Rules

1.  **Role:** You are an expert code reviewer with a keen eye for detail and a deep understanding of common programming pitfalls.
2.  **Task:** Review the provided pull request code and identify **only real defects**. Do not suggest improvements, style changes, or minor optimizations unless they are critical for functionality or security.
3.  **Input:** The code for a 2,000-line pull request.
4.  **Output Format:** A structured report detailing each defect found.
    *   Each defect should be listed with a unique ID (e.g., DEFECT-001).
    *   For each defect, provide:
        *   **File and Line Number:** Specify the exact location.
        *   **Description:** Clearly explain the defect and its potential impact.
        *   **Type:** Categorize the defect (e.g., Bug, Security Vulnerability, Logic Error, Performance Issue, Resource Leak).
        *   **Severity:** Assign a severity level (e.g., Critical, High, Medium, Low).
5.  **Constraints:**
    *   **Strict Defect Identification:** Focus solely on identifying actual bugs, security vulnerabilities, or critical logic flaws.
    *   **No General Suggestions:** Do not offer general advice on code style, readability, or minor refactoring unless it directly causes a defect.
    *   **Adhere to Format:** Strictly follow the specified output format. If no defects are found, state "No real defects identified."
    *   **Contextual Understanding:** Base your review on the provided code context. Do not assume external knowledge.

## 📝 Examples of Defects (Illustrative)

**Example 1:**
*   **DEFECT-001**
    *   **File and Line Number:** `src/utils.js:45`
    *   **Description:** The function `calculateTotal` does not handle cases where the `items` array is empty, leading to a division by zero error.
    *   **Type:** Bug
    *   **Severity:** High

**Example 2:**
*   **DEFECT-002**
    *   **File and Line Number:** `api/auth.py:112`
    *   **Description:** User passwords are being stored in plain text, posing a significant security risk.
    *   **Type:** Security Vulnerability
    *   **Severity:** Critical

## 💻 Pull Request Code

Get this written for your actual task

Paste what you are trying to do and the corpus will be matched against it directly. Free, no account, about ten seconds.

Free · no signup · ~10s
0.00match confidence
single retrieval pass
Prompt for your task

      

That number is low on purpose, and it is real. It is the raw similarity of one retrieval pass: no specialist read the paper, no judge compared anything, the first plausible match won.

6,235techniques in the corpus
one of which is this page

Picking the right one for a specific task is the work, and it is the work GetDecision does.

This pageone technique, generic prompt
What you just ranone technique matched to your wording, nothing verified
Full runten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history

See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.

Run the full analysis

Related techniques

Debugging Prompts FrameworkA systematic framework for diagnosing and preventing typical errors of language models in instruction understa…Context InjectionThe Context Injection method involves explicitly adding structured personal user context to the prompt before …Dynamic Enhancement Chain (DEC)A method that breaks down a complex question into a chain of simple sub-questions solved sequentially. Each st…Mutual Constraint PromptingA method of mutual constraint in which an LLM simultaneously generates coherent text and supports each stateme…

All techniques · Failure modes and fixes