SDK Reference

A complete guide to the Python SDK, covering all available classes, methods, usage patterns, and configuration options.

Project Class

The Project class is the main entry point for interacting with a Prompt Lockbox project via the Python SDK. It represents your entire prompt library and provides methods for finding, creating, and managing all prompts within it.

Initialization

Project(path=None)

Initializes the Project, finding its root directory and loading its configuration.

Parameters

path · str | Path | None

An optional path to a directory within the project. If None, it searches upwards from the current working directory to find the project root.

Raises

FileNotFoundError · exception

If no plb.toml file is found, indicating it’s not a valid project.

Example

from prompt_lockbox.api import Project

# Initialize from anywhere inside the project
project = Project()

Properties

root

The root pathlib.Path object of the Prompt Lockbox project.

Returns

root · pathlib.Path

The absolute path to the project’s root directory.

Methods

get_prompt(identifier)

Finds a single prompt by its name, ID, or file path. If a name is provided, it returns the Prompt object for the latest version.

Parameters

identifier · string · required

The name, ID, or file path string of the prompt to find.

Returns

prompt · Prompt | None

A Prompt object if found, otherwise None.

list_prompts()

Returns a list of all prompts found in the project.

Returns

prompts · List[Prompt]

A list of Prompt objects for every valid prompt file found.

create_prompt(…)

Creates a new prompt file on disk from the given metadata.

name · string · required

The name of the new prompt.

version · string · default: "1.0.0"

The starting semantic version for the prompt.

author · str | None

The author of the prompt. If not provided, it attempts to use the current Git user.

description · string

A short, human-readable summary of the prompt’s purpose.

namespace · str | List[str] | None

A list of strings to organize the prompt in a hierarchy (e.g., [‘billing’, ‘invoices’]).

tags · List[str] | None

A list of lowercase keywords for search and discovery.

intended_model · str

The specific LLM this prompt is designed for (e.g., openai/gpt-4o-mini).

notes · str

Any extra comments, warnings, or usage instructions.

model_parameters · Dict[str, Any] | None

A dictionary of model parameters to be stored.

linked_prompts · List[str] | None

A list of IDs of other prompts that are related to this one.

Returns

new_prompt · Prompt

A Prompt object representing the newly created file.

get_prompt(identifier)

Finds a single prompt by its name, ID, or file path. If a name is provided, it returns the Prompt object for the latest version.

Parameters

identifier · string · required

The name, ID, or file path string of the prompt to find.

Returns

prompt · string · required

A Prompt object if found, otherwise None.

search(…)

Searches for prompts using a specified method.

Parameters

query · str · required

The search query string.

method · string · default: "fuzzy"

The search method to use. Choices: fuzzy, hybrid, splade.

limit · int · default: "10"

The maximum number of results to return.

alpha · float

(Hybrid search only) A float between 0.0 (keyword) and 1.0 (semantic) to balance the search.

Returns

results · list[dict]

A list of result dictionaries, sorted by relevance.

index(method=‘hybrid’)

Builds a search index for all prompts to enable advanced search.

method · str · default: "hybrid"

The indexing method to use. Choices: hybrid or splade.

lint()

Validates all prompt files in the project for correctness and consistency.

Returns

report · dict

A dictionary of results categorized by check type.

get_status_report()

Generates a report of the lock status of all prompts.

Returns

report · dict

A dictionary categorizing prompts into locked, unlocked, tampered, and missing.

document_all(prompts_to_document=None)

Uses an AI to automatically document a list of prompts, or all prompts if none are provided.

Parameters

prompts_to_document · List[Prompt] | None

A specific list of Prompt objects to document. If None, all prompts in the project will be processed.

get_ai_config()

Retrieves the AI configuration from the project’s plb.toml file.

Returns

config · dict

A dictionary with provider and model keys.

Prompt Class

The Prompt class represents a single, versioned prompt file. It is the primary object you’ll work with to render, validate, and manage an individual prompt.

Properties

These are read-only attributes that provide quick access to the prompt’s data.

path

The absolute pathlib.Path to the prompt’s .yml file.

data

A dict containing all the parsed data from the YAML file.

name

The name of the prompt as a string.

version

The version of the prompt as a string (e.g., “1.0.0”).

description

The description of the prompt as a string.

required_variables

A set of all undeclared template variables found in the template string (e.g., ).

Methods

render(strict=True, **kwargs)

Parameters

strict · bool · default: "True"

If True, the method will raise an UndefinedError for any missing variables. If False, it will render missing variables as <<variable_name>> in the output string.

**kwargs · any

The key-value pairs to inject into the template. These will override any default_inputs specified in the prompt file.

Returns

rendered_text · str

The final, rendered prompt text as a string.

run(kwargs)

Renders the prompt, calls the configured LLM, and returns a structured result.

Parameters

result · dict

The key-value pairs to inject into the template before sending the prompt to the LLM.

Returns

rendered_text · str

A dictionary containing the rendered prompt, LLM output, model details, and usage statistics.

execute

Renders the prompt with given variables and executes it against the configured AI provider, returning the live response. This is the primary method for getting an AI response from a prompt.

Parameters

**kwargs · Any

The variables to inject into the template before execution, provided as keyword arguments (e.g., customer_name=“Jane Doe”).

Returns

rendered_text · str | dict

The return type depends on whether an output_schema is defined in your prompt’s .yml file.

• If no output_schema is present (default): It returns a str containing the raw text response from the language model.
• If an output_schema is present: It returns a dict containing the structured, parsed data that conforms to your schema.

lock()

Creates a lock entry for this prompt in the project’s lockfile. This records the file’s current SHA256 hash and a timestamp, marking it as secure.

unlock()

Removes the lock entry for this prompt from the project’s lockfile, allowing it to be edited.

verify()

Verifies the integrity of this prompt against the lockfile.

Returns

verification_status · tuple[bool, str]

A tuple containing a boolean (True if secure) and a status string (‘OK’, ‘UNLOCKED’, ‘TAMPERED’).

new_version(bump_type=‘minor’, author=None)

Creates a new, version-bumped copy of this prompt file and returns a Prompt object for the new file.

Parameters

bump_type · str · default: "minor"

The type of version bump to perform. Choices: major, minor, patch.

author · str | None

The author for the new version. If None, it defaults to the author of the source prompt or the current Git user.

Returns

new_prompt · Prompt

A new Prompt object representing the newly created file.

document()

Uses an AI to analyze the prompt’s template and automatically generate and save a description and tags for it. The original file’s comments and layout are preserved.

get_critique(note=…)

Gets an AI-powered critique and suggestions for improving the prompt. This method does not modify the file.

Parameters

note · str

A specific instruction for the AI on how to improve the prompt (e.g., “Make it more robust”).

Returns

critique_data · dict

A dictionary containing the ‘critique’, ‘suggestions’, and ‘improved_template’.

improve(improved_template)

Overwrites the prompt’s template block with a new version and updates its last_update timestamp. The original file’s comments and layout are preserved.

Parameters

improved_template · str · required

The new template string to write to the file. This is typically sourced from the result of .get_critique().

Congo! Explore more.

Assistant

Responses are generated using AI and may contain mistakes.