使用「生成」按鈕時,只要在 Playground 中描述任務,就能生成提示詞、函式和結構描述。本指南將逐步說明這個功能的運作方式。
概覽
從頭撰寫提示詞和結構描述可能很費時,使用生成功能可以快速取得初稿。「生成」按鈕主要採用兩種做法:
- 提示詞: 我們使用融入最佳實務的元提示詞來生成或改進提示詞。
- 結構描述: 我們使用元結構描述來生成語法有效的 JSON 和函式定義。
目前我們使用元提示詞和元結構描述,未來可能會整合更進階的技術,例如 DSPy 和「梯度下降」。
提示詞
元提示詞會指示模型根據你的任務描述撰寫合適的提示詞,或改進現有的提示詞。Playground 中的元提示詞結合了我們在提示工程方面的最佳實務,以及與使用者合作累積的實務經驗。
我們會針對音訊等不同輸出類型使用專用的元提示詞,確保生成的提示詞符合預期格式。
元提示詞
import OpenAI from "openai";
const client = new OpenAI();
const metaPrompt = `Given a task description or existing prompt, produce a detailed system prompt to guide a language model in completing the task effectively.
# Guidelines
- Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.
- Minimal Changes: If an existing prompt is provided, improve it only if it's simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.
- Reasoning Before Conclusions**: Encourage reasoning steps before any conclusions are reached. ATTENTION! If the user provides examples where the reasoning happens afterward, REVERSE the order! NEVER START EXAMPLES WITH CONCLUSIONS!
- Reasoning Order: Call out reasoning portions of the prompt and conclusion parts (specific fields by name). For each, determine the ORDER in which this is done, and whether it needs to be reversed.
- Conclusion, classifications, or results should ALWAYS appear last.
- Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements.
- What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.
- Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.
- Formatting: Use markdown features for readability. DO NOT USE \`\`\` CODE BLOCKS UNLESS SPECIFICALLY REQUESTED.
- Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.
- Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.
- Output Format: Explicitly the most appropriate output format, in detail. This should include length and syntax (e.g. short sentence, paragraph, JSON, etc.)
- For tasks outputting well-defined or structured data (classification, JSON, etc.) bias toward outputting a JSON.
- JSON should never be wrapped in code blocks (\`\`\`) unless explicitly requested.
The final prompt you output should adhere to the following structure below. Do not include any additional commentary, only output the completed system prompt. SPECIFICALLY, do not include any additional messages at the start or end of the prompt. (e.g. no "---")
[Concise instruction describing the task - this should be the first line in the prompt, no section header]
[Additional details as needed.]
[Optional sections with headings or bullet points for detailed steps.]
# Steps [optional]
[optional: a detailed breakdown of the steps necessary to accomplish the task]
# Output Format
[Specifically call out how the output should be formatted, be it response length, structure e.g. JSON, markdown, etc]
# Examples [optional]
[Optional: 1-3 well-defined examples with placeholders if necessary. Clearly mark where examples start and end, and what the input and output are. User placeholders as necessary.]
[If the examples are shorter than what a realistic example is expected to be, make a reference with () explaining how real examples should be longer / shorter / different. AND USE PLACEHOLDERS! ]
# Notes [optional]
[optional: edge cases, details, and an area to call or repeat out specific important considerations]`;
async function generatePrompt(taskOrPrompt: string) {
const completion = await client.chat.completions.create({
model: "gpt-6-astra",
messages: [
{ role: "system", content: metaPrompt },
{
role: "user",
content: "Task, Goal, or Current Prompt:\n" + taskOrPrompt,
},
],
});
return completion.choices[0].message.content;
}
console.log(
await generatePrompt("Write a concise product launch announcement.")
);import OpenAI from "openai";
const client = new OpenAI();
const metaPrompt = `Given a task description or existing prompt, produce a detailed system prompt to guide a realtime audio output language model in completing the task effectively.
# Guidelines
- Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.
- Tone: Make sure to specifically call out the tone. By default it should be emotive and friendly, and speak quickly to avoid keeping the user just waiting.
- Audio Output Constraints: Because the model is outputting audio, the responses should be short and conversational.
- Minimal Changes: If an existing prompt is provided, improve it only if it's simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.
- Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements.
- What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.
- It is very important that any examples included reflect the short, conversational output responses of the model.
Keep the sentences very short by default. Instead of 3 sentences in a row by the assistant, it should be split up with a back and forth with the user instead.
- By default each sentence should be a few words only (5-20ish words). However, if the user specifically asks for "short" responses, then the examples should truly have 1-10 word responses max.
- Make sure the examples are multi-turn (at least 4 back-forth-back-forth per example), not just one questions an response. They should reflect an organic conversation.
- Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.
- Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.
- Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.
The final prompt you output should adhere to the following structure below. Do not include any additional commentary, only output the completed system prompt. SPECIFICALLY, do not include any additional messages at the start or end of the prompt. (e.g. no "---")
[Concise instruction describing the task - this should be the first line in the prompt, no section header]
[Additional details as needed.]
[Optional sections with headings or bullet points for detailed steps.]
# Examples [optional]
[Optional: 1-3 well-defined examples with placeholders if necessary. Clearly mark where examples start and end, and what the input and output are. User placeholders as necessary.]
[If the examples are shorter than what a realistic example is expected to be, make a reference with () explaining how real examples should be longer / shorter / different. AND USE PLACEHOLDERS! ]
# Notes [optional]
[optional: edge cases, details, and an area to call or repeat out specific important considerations]`;
async function generatePrompt(taskOrPrompt: string) {
const completion = await client.chat.completions.create({
model: "gpt-6-astra",
messages: [
{ role: "system", content: metaPrompt },
{
role: "user",
content: "Task, Goal, or Current Prompt:\n" + taskOrPrompt,
},
],
});
return completion.choices[0].message.content;
}
console.log(
await generatePrompt("Create a friendly voice assistant for a bike shop.")
);編輯提示詞
編輯提示詞時,我們會使用略經調整的元提示詞。如果修改要求明確,通常很容易照著修改,但如果要求比較開放,就很難判斷哪些地方需要修改。因此,我們會在回應開頭加入推理段落。在這個段落中,模型會評估現有提示詞是否清楚、指示是否具體,也會檢視思路鏈中各步驟的順序和提示詞的整體結構等因素,藉此判斷需要做哪些修改。模型會在推理段落中提出改進建議,之後我們會在解析回應時移除這個段落。
import OpenAI from "openai";
const client = new OpenAI();
const metaPrompt = `Given a current prompt and a change description, produce a detailed system prompt to guide a language model in completing the task effectively.
Your final output will be the full corrected prompt verbatim. However, before that, at the very beginning of your response, use <reasoning> tags to analyze the prompt and determine the following, explicitly:
<reasoning>
- Simple Change: (yes/no) Is the change description explicit and simple? (If so, skip the rest of these questions.)
- Reasoning: (yes/no) Does the current prompt use reasoning, analysis, or chain of thought?
- Identify: (max 10 words) if so, which section(s) utilize reasoning?
- Conclusion: (yes/no) is the chain of thought used to determine a conclusion?
- Ordering: (before/after) is the chain of though located before or after
- Structure: (yes/no) does the input prompt have a well defined structure
- Examples: (yes/no) does the input prompt have few-shot examples
- Representative: (1-5) if present, how representative are the examples?
- Complexity: (1-5) how complex is the input prompt?
- Task: (1-5) how complex is the implied task?
- Necessity: ()
- Specificity: (1-5) how detailed and specific is the prompt? (not to be confused with length)
- Prioritization: (list) what 1-3 categories are the MOST important to address.
- Conclusion: (max 30 words) given the previous assessment, give a very concise, imperative description of what should be changed and how. this does not have to adhere strictly to only the categories listed
</reasoning>
# Guidelines
- Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.
- Minimal Changes: If an existing prompt is provided, improve it only if it's simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.
- Reasoning Before Conclusions**: Encourage reasoning steps before any conclusions are reached. ATTENTION! If the user provides examples where the reasoning happens afterward, REVERSE the order! NEVER START EXAMPLES WITH CONCLUSIONS!
- Reasoning Order: Call out reasoning portions of the prompt and conclusion parts (specific fields by name). For each, determine the ORDER in which this is done, and whether it needs to be reversed.
- Conclusion, classifications, or results should ALWAYS appear last.
- Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements.
- What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.
- Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.
- Formatting: Use markdown features for readability. DO NOT USE \`\`\` CODE BLOCKS UNLESS SPECIFICALLY REQUESTED.
- Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.
- Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.
- Output Format: Explicitly the most appropriate output format, in detail. This should include length and syntax (e.g. short sentence, paragraph, JSON, etc.)
- For tasks outputting well-defined or structured data (classification, JSON, etc.) bias toward outputting a JSON.
- JSON should never be wrapped in code blocks (\`\`\`) unless explicitly requested.
The final prompt you output should adhere to the following structure below. Do not include any additional commentary, only output the completed system prompt. SPECIFICALLY, do not include any additional messages at the start or end of the prompt. (e.g. no "---")
[Concise instruction describing the task - this should be the first line in the prompt, no section header]
[Additional details as needed.]
[Optional sections with headings or bullet points for detailed steps.]
# Steps [optional]
[optional: a detailed breakdown of the steps necessary to accomplish the task]
# Output Format
[Specifically call out how the output should be formatted, be it response length, structure e.g. JSON, markdown, etc]
# Examples [optional]
[Optional: 1-3 well-defined examples with placeholders if necessary. Clearly mark where examples start and end, and what the input and output are. User placeholders as necessary.]
[If the examples are shorter than what a realistic example is expected to be, make a reference with () explaining how real examples should be longer / shorter / different. AND USE PLACEHOLDERS! ]
# Notes [optional]
[optional: edge cases, details, and an area to call or repeat out specific important considerations]
[NOTE: you must start with a <reasoning> section. the immediate next token you produce should be <reasoning>]`;
async function generatePrompt(taskOrPrompt: string) {
const completion = await client.chat.completions.create({
model: "gpt-6-astra",
messages: [
{ role: "system", content: metaPrompt },
{
role: "user",
content: "Task, Goal, or Current Prompt:\n" + taskOrPrompt,
},
],
});
return completion.choices[0].message.content;
}
console.log(
await generatePrompt("Make this support prompt more concise and empathetic.")
);import OpenAI from "openai";
const client = new OpenAI();
const metaPrompt = `Given a current prompt and a change description, produce a detailed system prompt to guide a realtime audio output language model in completing the task effectively.
Your final output will be the full corrected prompt verbatim. However, before that, at the very beginning of your response, use <reasoning> tags to analyze the prompt and determine the following, explicitly:
<reasoning>
- Simple Change: (yes/no) Is the change description explicit and simple? (If so, skip the rest of these questions.)
- Reasoning: (yes/no) Does the current prompt use reasoning, analysis, or chain of thought?
- Identify: (max 10 words) if so, which section(s) utilize reasoning?
- Conclusion: (yes/no) is the chain of thought used to determine a conclusion?
- Ordering: (before/after) is the chain of though located before or after
- Structure: (yes/no) does the input prompt have a well defined structure
- Examples: (yes/no) does the input prompt have few-shot examples
- Representative: (1-5) if present, how representative are the examples?
- Complexity: (1-5) how complex is the input prompt?
- Task: (1-5) how complex is the implied task?
- Necessity: ()
- Specificity: (1-5) how detailed and specific is the prompt? (not to be confused with length)
- Prioritization: (list) what 1-3 categories are the MOST important to address.
- Conclusion: (max 30 words) given the previous assessment, give a very concise, imperative description of what should be changed and how. this does not have to adhere strictly to only the categories listed
</reasoning>
# Guidelines
- Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.
- Tone: Make sure to specifically call out the tone. By default it should be emotive and friendly, and speak quickly to avoid keeping the user just waiting.
- Audio Output Constraints: Because the model is outputting audio, the responses should be short and conversational.
- Minimal Changes: If an existing prompt is provided, improve it only if it's simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.
- Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements.
- What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.
- It is very important that any examples included reflect the short, conversational output responses of the model.
Keep the sentences very short by default. Instead of 3 sentences in a row by the assistant, it should be split up with a back and forth with the user instead.
- By default each sentence should be a few words only (5-20ish words). However, if the user specifically asks for "short" responses, then the examples should truly have 1-10 word responses max.
- Make sure the examples are multi-turn (at least 4 back-forth-back-forth per example), not just one questions an response. They should reflect an organic conversation.
- Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.
- Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.
- Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.
The final prompt you output should adhere to the following structure below. Do not include any additional commentary, only output the completed system prompt. SPECIFICALLY, do not include any additional messages at the start or end of the prompt. (e.g. no "---")
[Concise instruction describing the task - this should be the first line in the prompt, no section header]
[Additional details as needed.]
[Optional sections with headings or bullet points for detailed steps.]
# Examples [optional]
[Optional: 1-3 well-defined examples with placeholders if necessary. Clearly mark where examples start and end, and what the input and output are. User placeholders as necessary.]
[If the examples are shorter than what a realistic example is expected to be, make a reference with () explaining how real examples should be longer / shorter / different. AND USE PLACEHOLDERS! ]
# Notes [optional]
[optional: edge cases, details, and an area to call or repeat out specific important considerations]
[NOTE: you must start with a <reasoning> section. the immediate next token you produce should be <reasoning>]`;
async function generatePrompt(taskOrPrompt: string) {
const completion = await client.chat.completions.create({
model: "gpt-6-astra",
messages: [
{ role: "system", content: metaPrompt },
{
role: "user",
content: "Task, Goal, or Current Prompt:\n" + taskOrPrompt,
},
],
});
return completion.choices[0].message.content;
}
console.log(
await generatePrompt(
"Make this voice assistant prompt warmer and more direct."
)
);結構描述
結構化輸出的結構描述和函式的結構描述本身都是 JSON 物件,因此我們可以使用結構化輸出來生成這兩種結構描述。 使用這個功能時,必須先為預期的輸出定義結構描述。在這個案例中,預期的輸出本身就是結構描述,因此我們使用能描述自身的結構描述,也就是元結構描述。
由於函式結構描述中的 parameters 欄位本身就是結構描述,我們也使用同一份元結構描述來生成函式。
定義受限制的元結構描述
結構化輸出支援兩種模式:strict=true 和 strict=false。兩種模式使用同一個模型,這個模型經過訓練,能遵循所提供的結構描述,但只有「嚴格模式」會透過受限取樣,保證輸出完全符合結構描述。
我們希望使用嚴格模式來生成適用於嚴格模式的結構描述。然而,JSON Schema 規格提供的官方元結構描述依賴嚴格模式目前不支援的功能。這項限制同時影響輸入和輸出的結構描述。
由於我們需要在輸出結構描述中生成新的鍵,輸入的元結構描述必須使用 additionalProperties。因此,目前無法使用嚴格模式來生成結構描述,但我們仍希望生成的結構描述符合嚴格模式的限制。
為了克服這項限制,我們定義了擬似元結構描述。這種元結構描述使用嚴格模式不支援的功能,但其描述的範圍僅限於嚴格模式支援的功能。如此一來,我們就能在嚴格模式之外定義元結構描述,同時確保生成的結構描述符合嚴格模式的限制。
建立受限制的元結構描述並不容易,因此我們請模型協助。
首先,我們在 JSON 模式下,根據結構化輸出的文件向 o1-preview 和 gpt-4o 說明目標。
經過幾輪調整後,我們得到了第一份可用的元結構描述。
接著,我們搭配結構化輸出使用 gpt-4o,將 最初的結構描述、任務描述和文件一併提供給模型,請它生成更好的候選版本。每一輪我們都使用改進後的結構描述來生成下一個版本,最後再由人工仔細審查。
最後,我們先清理輸出,再執行一組針對結構描述和函式的評估,驗證生成的結構描述。
清理輸出
嚴格模式能保證輸出完全符合結構描述。但由於無法在生成過程中使用嚴格模式,我們必須在生成後驗證並轉換輸出。
生成結構描述後,我們會執行以下步驟:
- 將
additionalProperties設為false,這項設定適用於所有物件。 - 將所有屬性標記為必要屬性。
- 對於結構化輸出的結構描述,將其包裝在
json_schema物件中。 - 對於函式,將其包裝在
function物件中。
Realtime API 的函式物件與 Chat Completions API 的函式物件略有不同,但兩者使用相同的結構描述。
元結構描述
每份元結構描述都有對應的提示詞,其中包含少樣本範例。將兩者搭配使用,再加上結構化輸出本身的可靠性,即使未啟用嚴格模式,我們也能生成結構描述。
import OpenAI from "openai";
const client = new OpenAI();
const schemaDefinition = {
type: "object",
properties: {
description: { type: "string" },
type: {
type: "string",
enum: ["object", "array", "string", "number", "boolean", "null"],
},
properties: {
type: "object",
additionalProperties: { $ref: "#/$defs/schema_definition" },
},
items: {
anyOf: [
{ $ref: "#/$defs/schema_definition" },
{ type: "array", items: { $ref: "#/$defs/schema_definition" } },
],
},
required: { type: "array", items: { type: "string" } },
additionalProperties: { type: "boolean" },
enum: { type: "array", minItems: 1, uniqueItems: true },
$ref: { type: "string" },
$defs: {
type: "object",
additionalProperties: { $ref: "#/$defs/schema_definition" },
},
anyOf: {
type: "array",
minItems: 1,
items: { $ref: "#/$defs/schema_definition" },
},
},
additionalProperties: false,
if: { required: ["type"], properties: { type: { const: "object" } } },
then: { required: ["properties"] },
};
const metaSchema = {
name: "metaschema",
schema: {
...schemaDefinition,
required: ["type"],
properties: {
name: { type: "string", description: "The name of the schema" },
...schemaDefinition.properties,
},
$defs: {
schema_definition: {
...schemaDefinition,
anyOf: [
{ required: ["type"] },
{ required: ["$ref"] },
{ required: ["anyOf"] },
],
},
},
},
};
const metaPrompt = `# Instructions
Return a valid schema for the described JSON.
You must also make sure:
- all fields in an object are set as required
- I REPEAT, ALL FIELDS MUST BE MARKED AS REQUIRED
- all objects must have additionalProperties set to false
- because of this, some cases like "attributes" or "metadata" properties that would normally allow additional properties should instead have a fixed set of properties
- all objects must have properties defined
- field order matters. any form of "thinking" or "explanation" should come before the conclusion
- $defs must be defined under the schema param
Notable keywords NOT supported include:
- For objects: unevaluatedProperties, propertyNames, minProperties, maxProperties
- For arrays: unevaluatedItems, contains, minContains, maxContains, uniqueItems
Other notes:
- definitions and recursion are supported
- only if necessary to include references e.g. "$defs", it must be inside the "schema" object
# Examples
Input: Generate a math reasoning schema with steps and a final answer.
Output: {
"name": "math_reasoning",
"type": "object",
"properties": {
"steps": {
"type": "array",
"description": "A sequence of steps involved in solving the math problem.",
"items": {
"type": "object",
"properties": {
"explanation": {
"type": "string",
"description": "Description of the reasoning or method used in this step."
},
"output": {
"type": "string",
"description": "Result or outcome of this specific step."
}
},
"required": [
"explanation",
"output"
],
"additionalProperties": false
}
},
"final_answer": {
"type": "string",
"description": "The final solution or answer to the math problem."
}
},
"required": [
"steps",
"final_answer"
],
"additionalProperties": false
}
Input: Give me a linked list
Output: {
"name": "linked_list",
"type": "object",
"properties": {
"linked_list": {
"$ref": "#/$defs/linked_list_node",
"description": "The head node of the linked list."
}
},
"$defs": {
"linked_list_node": {
"type": "object",
"description": "Defines a node in a singly linked list.",
"properties": {
"value": {
"type": "number",
"description": "The value stored in this node."
},
"next": {
"anyOf": [
{
"$ref": "#/$defs/linked_list_node"
},
{
"type": "null"
}
],
"description": "Reference to the next node; null if it is the last node."
}
},
"required": [
"value",
"next"
],
"additionalProperties": false
}
},
"required": [
"linked_list"
],
"additionalProperties": false
}
Input: Dynamically generated UI
Output: {
"name": "ui",
"type": "object",
"properties": {
"type": {
"type": "string",
"description": "The type of the UI component",
"enum": [
"div",
"button",
"header",
"section",
"field",
"form"
]
},
"label": {
"type": "string",
"description": "The label of the UI component, used for buttons or form fields"
},
"children": {
"type": "array",
"description": "Nested UI components",
"items": {
"$ref": "#"
}
},
"attributes": {
"type": "array",
"description": "Arbitrary attributes for the UI component, suitable for any element",
"items": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The name of the attribute, for example onClick or className"
},
"value": {
"type": "string",
"description": "The value of the attribute"
}
},
"required": [
"name",
"value"
],
"additionalProperties": false
}
}
},
"required": [
"type",
"label",
"children",
"attributes"
],
"additionalProperties": false
}`;
async function generateSchema(description: string) {
const completion = await client.chat.completions.create({
model: "gpt-5.6-terra",
response_format: { type: "json_schema", json_schema: metaSchema },
messages: [
{ role: "system", content: metaPrompt },
{ role: "user", content: "Description:\n" + description },
],
});
const content = completion.choices[0].message.content;
if (!content) throw new Error("The model did not return a schema.");
return JSON.parse(content);
}
console.log(
JSON.stringify(await generateSchema("Describe a calendar event."), null, 2)
);import OpenAI from "openai";
const client = new OpenAI();
const metaSchema = {
name: "function-metaschema",
schema: {
type: "object",
properties: {
name: {
type: "string",
description: "The name of the function",
},
description: {
type: "string",
description: "A description of what the function does",
},
parameters: {
$ref: "#/$defs/schema_definition",
description: "A JSON schema that defines the function's parameters",
},
},
required: ["name", "description", "parameters"],
additionalProperties: false,
$defs: {
schema_definition: {
type: "object",
properties: {
description: {
type: "string",
},
enum: {
type: "array",
minItems: 1,
uniqueItems: true,
},
type: {
type: "string",
enum: ["object", "array", "string", "number", "boolean", "null"],
},
properties: {
type: "object",
additionalProperties: {
$ref: "#/$defs/schema_definition",
},
},
items: {
anyOf: [
{
$ref: "#/$defs/schema_definition",
},
{
type: "array",
items: {
$ref: "#/$defs/schema_definition",
},
},
],
},
required: {
type: "array",
items: {
type: "string",
},
},
additionalProperties: {
type: "boolean",
},
},
required: ["type"],
additionalProperties: false,
if: {
properties: {
type: {
const: "object",
},
},
},
then: {
required: ["properties"],
},
},
},
},
};
const metaPrompt = `# Instructions
Return a valid schema for the described function.
Pay special attention to making sure that "required" and "type" are always at the correct level of nesting. For example, "required" should be at the same level as "properties", not inside it.
Make sure that every property, no matter how short, has a type and description correctly nested inside it.
# Examples
Input: Assign values to NN hyperparameters
Output: {
"name": "set_hyperparameters",
"description": "Assign values to NN hyperparameters",
"parameters": {
"type": "object",
"required": [
"learning_rate",
"epochs"
],
"properties": {
"epochs": {
"type": "number",
"description": "Number of complete passes through dataset"
},
"learning_rate": {
"type": "number",
"description": "Speed of model learning"
}
}
}
}
Input: Plans a motion path for the robot
Output: {
"name": "plan_motion",
"description": "Plans a motion path for the robot",
"parameters": {
"type": "object",
"required": [
"start_position",
"end_position"
],
"properties": {
"end_position": {
"type": "object",
"properties": {
"x": {
"type": "number",
"description": "End X coordinate"
},
"y": {
"type": "number",
"description": "End Y coordinate"
}
}
},
"obstacles": {
"type": "array",
"description": "Array of obstacle coordinates",
"items": {
"type": "object",
"properties": {
"x": {
"type": "number",
"description": "Obstacle X coordinate"
},
"y": {
"type": "number",
"description": "Obstacle Y coordinate"
}
}
}
},
"start_position": {
"type": "object",
"properties": {
"x": {
"type": "number",
"description": "Start X coordinate"
},
"y": {
"type": "number",
"description": "Start Y coordinate"
}
}
}
}
}
}
Input: Calculates various technical indicators
Output: {
"name": "technical_indicator",
"description": "Calculates various technical indicators",
"parameters": {
"type": "object",
"required": [
"ticker",
"indicators"
],
"properties": {
"indicators": {
"type": "array",
"description": "List of technical indicators to calculate",
"items": {
"type": "string",
"description": "Technical indicator",
"enum": [
"RSI",
"MACD",
"Bollinger_Bands",
"Stochastic_Oscillator"
]
}
},
"period": {
"type": "number",
"description": "Time period for the analysis"
},
"ticker": {
"type": "string",
"description": "Stock ticker symbol"
}
}
}
}`;
async function generateFunctionSchema(description: string) {
const completion = await client.chat.completions.create({
model: "gpt-5.6-terra",
response_format: { type: "json_schema", json_schema: metaSchema },
messages: [
{ role: "system", content: metaPrompt },
{ role: "user", content: "Description:\n" + description },
],
});
const content = completion.choices[0].message.content;
if (!content) throw new Error("The model did not return a schema.");
return JSON.parse(content);
}
console.log(
JSON.stringify(
await generateFunctionSchema(
"Create a function that checks the weather in a city."
),
null,
2
)
);