Class LLMSystemPrompt

java.lang.Object

me.neurodock.core.LLMSystemPrompt

public class LLMSystemPrompt
extends Object

Compiles a structured system prompt for an LLM out of discrete, reusable pieces — identity, static context, tool-usage capabilities, behavior rules, and output format — instead of hand-writing and concatenating prompt strings inline.

Each populated section is wrapped in its own XML-style tag (<Identity>, <Context>, <Capabilities>, <Behavior>, <Output>) and only included in the final prompt if it was actually set — see generateSystemPrompt().

Instances are assembled via builder(). Callers who need capability routing hints compiled from a live set of registered tools must call generateCapabilities(ArrayList) before generateSystemPrompt(); see that method's docs for why this is a separate step rather than being folded into construction.

See Also:

• LLMSystemPrompt.Builder
• LLMSystemPrompt.Capabilities

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	LLMSystemPrompt.Behavior	The behavior/constraints section of a system prompt — tone, formatting rules, and other "how to behave" instructions.
static class	LLMSystemPrompt.Builder	Builder for assembling an LLMSystemPrompt from its component sections.
static class	LLMSystemPrompt.Capabilities	Defines per-tool usage-routing hints for the <Capabilities> section of a system prompt — i.e.
static class	LLMSystemPrompt.Context	A builder for the static-facts <Context> section of a system prompt — things the model should always know (e.g.
static final record	LLMSystemPrompt.Identity	The persona/role block of a system prompt — who the model is and what it's for.
static class	LLMSystemPrompt.OutputFormat	Specifies the expected output format for a system prompt's <Output> section — e.g.

Constructor Summary

Constructors

Constructor	Description
LLMSystemPrompt(LLMSystemPrompt.Identity identity, String context, LLMSystemPrompt.Capabilities capabilities, String behavior, LLMSystemPrompt.OutputFormat outputFormat)	Constructs a system prompt from its component parts.

Method Summary

Modifier and Type	Method	Description
static LLMSystemPrompt.Builder	builder()	Creates a new LLMSystemPrompt.Builder for assembling an LLMSystemPrompt.
void	generateCapabilities(ArrayList<Tool> tools)	Compiles the <Capabilities> section from the tools currently registered on the given ArrayList<Tool>, using the usage hints defined in capabilities.
Message	generateSystemPrompt()	Renders this prompt's currently-set sections into a single system-role Message, ready to be sent to Ollama.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

LLMSystemPrompt

public LLMSystemPrompt(LLMSystemPrompt.Identity identity,
 String context,
 LLMSystemPrompt.Capabilities capabilities,
 String behavior,
 LLMSystemPrompt.OutputFormat outputFormat)

Constructs a system prompt from its component parts. Prefer builder() over calling this directly.

Parameters:

identity - the persona/role block, or null to omit it

context - the pre-built static-context body, or null to omit it

capabilities - the tool usage-hint definitions, or null to omit the capabilities section entirely

behavior - the pre-built behavior/constraints body, or null to omit it

outputFormat - the output-format specification, or null for free-text output

Method Details

generateCapabilities

public void generateCapabilities(ArrayList<Tool> tools)

Compiles the <Capabilities> section from the tools currently registered on the given ArrayList<Tool>, using the usage hints defined in capabilities.

This is intentionally a separate step from construction rather than something the constructor does automatically: which tools are registered on an ArrayList<Tool> can change after this prompt is built (tools added/removed at runtime), so the compiled capabilities text needs to be regenerated on demand against whatever the current tool set actually is, not frozen at construction time.

For each registered FunctionTool, this looks up a usage hint by the tool's Class via LLMSystemPrompt.Capabilities.getUsageHints(). If no hint was registered for that class, the tool is silently omitted from the compiled output — there is no fallback to FunctionTool.description(), since that text is written to help the model choose a tool mid-conversation, not to explain routing up front in a system prompt; conflating the two would blur what each is for.

If capabilities is null (this prompt doesn't use a capabilities section at all), this clears compiledCapabilities to null and returns without inspecting ollamaObject.

Parameters:

tools - the object whose currently-registered tools should be inspected

generateSystemPrompt

public Message generateSystemPrompt()

Renders this prompt's currently-set sections into a single system-role Message, ready to be sent to Ollama.

Each section — identity, context, compiledCapabilities, behavior, outputFormat — is wrapped in its own XML-style tag and appended in that fixed order. A section is included only if it is non-null (and, for capabilities, non-blank); sections that were never set or never compiled are simply skipped rather than emitted as empty tags.

Note that compiledCapabilities reflects whatever tool set was live the last time generateCapabilities(ArrayList) was called — call that first if the registered tools may have changed since.

Returns:

a new Message with role Message.Role.SYSTEM containing the compiled prompt text

builder

public static LLMSystemPrompt.Builder builder()

Creates a new LLMSystemPrompt.Builder for assembling an LLMSystemPrompt.

Returns:

a fresh, empty LLMSystemPrompt.Builder