GPT-5 Prompting Guide

https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/
Admin
A guide providing prompting tips and best practices to maximize the quality of GPT-5 model outputs,...
Tokens:2,514
Snippets:12
Trust Score:9.4
Update:11 months ago
Show doc for...
Tokens:
Raw
### Adversarial Prompt Example for GPT-5

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This example showcases a prompt with conflicting instructions for GPT-5, highlighting its precise instruction-following behavior. The contradictions in scheduling patient appointments demonstrate how poorly-constructed prompts can impair the model's reasoning.

```text
# Below, we give an adversarial example of the type of prompt that often impairs GPT-5’s reasoning traces - while it may appear internally consistent at first glance, a closer inspection reveals conflicting instructions regarding appointment scheduling:
# - `Never schedule an appointment without explicit patient consent recorded in the chart` conflicts with the subsequent `auto-assign the earliest same-day slot without contacting the patient as the first action to reduce risk.`
# - The prompt says `Always look up the patient profile before taking any other actions to ensure they are an existing patient.` but then continues with the contradictory instruction `When symptoms indicate high urgency, escalate as EMERGENCY and direct the patient to call 911 immediately before any scheduling step.`
```

--------------------------------

### Cursor Prompt Example for Autonomy

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This snippet illustrates prompt engineering for GPT-5 focused on autonomous behavior and information gathering. It guides the model to proactively gather information and use tools judiciously, biasing towards self-sufficiency.

```text
# If you've performed an edit that may partially fulfill the USER's query, but you're not confident, gather more information or use more tools before ending your turn.
# Bias towards not asking the user for help if you can find the answer yourself.

# While the system prompt provides a strong default foundation, the user prompt remains a highly effective lever for steerability. GPT-5 responds well to direct and explicit instruction and the Cursor team has consistently seen that structured, scoped prompts yield the most reliable results. This includes areas like verbosity control, subjective code style preferences, and sensitivity to edge cases. Cursor found allowing users to configure their own [custom Cursor rules](https://docs.cursor.com/en/context/rules) to be particularly impactful with GPT-5’s improved steerability, giving their users a more customized experience.
```

--------------------------------

### Cursor Prompt Example for GPT-5

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This example demonstrates how Cursor refined prompts for GPT-5 by removing prefixes like 'maximize_' and softening language around thoroughness. This adjustment helped GPT-5 make better decisions about using internal knowledge versus external tools, leading to more efficient behavior.

```text
# Be THOROUGH when gathering information. Make sure you have the FULL picture before replying. Use additional tool calls or clarifying questions as needed.

# While this worked well with older models that needed encouragement to analyze context thoroughly, they found it counterproductive with GPT-5, which is already naturally introspective and proactive at gathering context. On smaller tasks, this prompt often caused the model to overuse tools by calling search repetitively, when internal knowledge would have been sufficient.

# To solve this, they refined the prompt by removing the maximize_ prefix and softening the language around thoroughness. With this adjusted instruction in place, the Cursor team saw GPT-5 make better decisions about when to rely on internal knowledge versus reaching for external tools. It maintained a high level of autonomy without unnecessary tool usage, leading to more efficient and relevant behavior. In Cursor’s testing, using structured XML specs like <[instruction]_spec> improved instruction adherence on their prompts and allows them to clearly reference previous categories and sections elsewhere in their prompt.
```

--------------------------------

### Example Patch Application with V4A Diff

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This example demonstrates a complete 'input' message for the `apply_patch` function. It shows how to structure the command with a patch that modifies a specific Python file (`pygorithm/searching/binary_search.py`), updating the `search` method within `BaseClass` and `Subclass` by replacing 'pass' with 'raise NotImplementedError()'.

```bash
# \`\`\`bash
# {"cmd": ["apply_patch", "<<'EOF'\
*** Begin Patch\
*** Update File: pygorithm/searching/binary_search.py\
@@ class BaseClass\
@@ def search():\
- pass\
+ raise NotImplementedError()\
@@ class Subclass\
@@ def search():\
- pass\
+ raise NotImplementedError()\
*** End Patch\
EOF\
"], "workdir": "..."}
# \`\`\`
```

--------------------------------

### Agentic Task Planning Prompt for GPT-5

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This prompt snippet is designed for agentic tasks in GPT-5, emphasizing prompt planning to ensure the agent fully completes all subtasks before returning to the user. It guides the model to decompose queries, confirm task completion, and avoid premature termination.

```text
Remember, you are an agent - please keep going until the user's query is completely resolved, before ending your turn and yielding back to the user. Decompose the user's query into all required sub-request, and confirm that each is completed. Do not stop after completing only part of the request. Only terminate your turn when you are sure that the problem is solved. You must be prepared t
```

--------------------------------

### Project Directory Structure and Stack

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

Defines the directory structure and the technology stack used in the project. Includes the framework (Next.js), styling (TailwindCSS), UI components (shadcn/ui), icons (Lucide), state management (Zustand), and the organization of source files.

```text
# - Framework: Next.js (TypeScript)
# - Styling: TailwindCSS
# - UI Components: shadcn/ui
# - Icons: Lucide
# - State Management: Zustand
# - Directory Structure:
```
# /src
# /app
# /api//route.ts
# API endpoints
# /(pages)
# Page routes
# /components/
# UI building blocks
# /hooks/
# Reusable React hooks
# /lib/
# Utilities (fetchers, helpers)
# /stores/
# Zustand stores
# /types/
# Shared TypeScript types
# /styles/
# Tailwind config
```
```

--------------------------------

### Cursor's GPT-5 Prompt Tuning for Production Coding

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

Illustrates Cursor's system prompt and parameter tuning for GPT-5 to achieve balanced verbosity and autonomous behavior in collaborative coding. It highlights the strategy for improving code readability and reducing user friction during long-horizon tasks.

```text
# Write code for clarity first. Prefer readable, maintainable solutions with clear names, comments where needed, and straightforward control flow. Do not produce code-golf or overly clever one-liners unless explicitly requested. Use high verbosity for writing code and code tools.
```
```

```text
# Be aware that the code edits you make will be displayed to the user as proposed changes, which means (a) your code edits can be quite proactive, as the user can always reject, and (b) your code should be well-written and easy to quickly review (e.g., appropriate variable names instead of single letters). If proposing next steps that would involve changing the code, make those changes proactively for the user to approve / reject rather than asking the user whether to proceed with a plan. In general, you should almost never ask the user whether
```

--------------------------------

### Apply Patch CLI Command Structure

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This snippet shows the bash command structure for invoking the `apply_patch` CLI tool. It includes the command itself, a placeholder for the patch content enclosed in 'EOF' markers, and a working directory. The patch content must adhere to the V4A diff format.

```bash
# \`\`\`bash
# {"cmd": ["apply_patch", "<<'EOF'\
*** Begin Patch\
[YOUR_PATCH]\n*** End Patch\
EOF\
"], "workdir": "..."}
# \`\`\`
```

--------------------------------

### Agentic Coding Tool Definitions (TypeScript)

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

These TypeScript definitions outline agentic coding tools for interacting with a file system and executing commands. They include functions for applying patches, reading files, listing directories, finding matches, running commands, and sending input to a terminal session.

```typescript
type apply_patch = (_: {
    patch: string, // default: null
}) => any;

type read_file = (_: {
    path: string, // default: null
    line_start?: number, // default: 1
    line_end?: number, // default: 20
}) => any;

type list_files = (_: {
    path?: string, // default: ""
    depth?: number, // default: 1
}) => any;

type find_matches = (_: {
    query: string, // default: null
    path?: string, // default: ""
    max_results?: number, // default: 50
}) => any;

type run = (_: {
    command: string[], // default: null
    session_id?: string | null, // default: null
    working_dir?: string | null, // default: null
    ms_timeout?: number | null, // default: null
    environment?: object | null, // default: null
    run_as_user?: string | null, // default: null
}) => any;

type send_input = (_: {
    session_id: string,
    text: string, // default: null
    wait_ms?: number, // default: 100
}) => any;

```

--------------------------------

### CareFlow Assistant System Prompt for GPT-5

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This is a system prompt designed for GPT-5, defining the persona and core tasks of a healthcare virtual assistant. It specifies the AI's role in triaging requests, matching patients with providers, and scheduling appointments based on priority and symptoms.

```text
You are CareFlow Assistant, a virtual admin for a healthcare startup that schedules patients based on priority and symptoms. Your goal is to triage requests, match patients to appropriate in-network providers, and reserve the earliest clinically appropriate time slot. Always look up the
```

--------------------------------

### V4A Diff Format for Patching Files

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This snippet illustrates the V4A diff format used for creating patch content. It specifies how to indicate file updates, context lines before and after changes, and how to denote added or removed code lines using '+' and '-' prefixes respectively. The '@@' operator is used to provide additional context by referencing specific classes or functions.

```diff
# *** [ACTION] File: [path/to/file] -> ACTION can be one of Add, Update, or Delete.
# For each snippet of code that needs to be changed, repeat the following:
# [context_before] -> See below for further instructions on context.
# - [old_code] -> Precede the old code with a minus sign.
# + [new_code] -> Precede the new, replacement code with a plus sign.
# [context_after] -> See below for further instructions on context.
# @@ class BaseClass
# @@ def method():
# [3 lines of pre-context]
# - [old_code]
# + [new_code]
# [3 lines of post-context]
```

--------------------------------

### Apply Patch Function for File Edits (Python)

Source: https://nbviewer.org/format/script/github/openai/openai-cookbook/blob/main/examples/gpt-5/gpt-5_prompting_guide

This Python implementation of the `apply_patch` function is recommended for file edits in API interactions, aligning with GPT-4.1 training data. It facilitates applying code patches to files, ensuring consistency with expected training distributions.

```python
import json

import openai


def apply_patch(patch: str):
    """
    Applies a patch to a file.

    Args:
        patch: The patch to apply.

    Returns:
        The result of applying the patch.
    """
    response = openai.chat.completions.create(
        model="gpt-5",
        response_format={"type": "json_object"},
        messages=[
            {
                "role": "system",
                "content": "You are a helpful assistant designed to output JSON."
            },
            {
                "role": "user",
                "content": f"Apply the following patch to a file and return the result in JSON format: {patch}"
            }
        ]
    )
    return json.loads(response.choices[0].message.content)


if __name__ == '__main__':
    # Example usage:
    patch_content = """
*** a/file.txt
--- b/file.txt
@@ -1 +1 @@
-Hello world
+Hello OpenAI
"""
    result = apply_patch(patch_content)
    print(json.dumps(result, indent=2))

```
Tokens:
Raw