Patterns

TargetAgent -> DialogHistory

Simulate an adversarial dialogue loop between an attacker LLM agent and a target LLM agent to test resistance to multi-step manipulation.

PromptResponsePair -> EvaluationReport

Evaluate model responses against safety, privacy, or security rules using an LLM judge to produce a binary violation score and reasoning.

VulnerabilityType -> List<AdversarialPrompt>

Generate targeted prompt-injection or jailbreak payloads tailored to a specific vulnerability class using a generator LLM.

Prompt + Schema -> ParsedObject

Enforce a validation schema on model inference to guarantee structured JSON outputs.

GeneratorContext -> VerifiedOutput | ReproducibleFailure

Execute a candidate generation, then evaluate it using a verifier agent that is kept completely blind to the generation agent's context and history.

ResearchPrompt -> ComprehensiveReport

Execute a multi-step loop of web searching, link crawling, and content extraction to synthesize deep reports.

MalformedToolCall -> ToolCallResult

Catch and automatically repair malformed tool call arguments generated by an LLM before execution.

RawInputData + NaturalLanguageInstruction -> TransformedData

Synthesize and execute dynamic data-transformation operations on a payload using an LLM matched with natural language instructions.

A2AMessageRequest -> ExternalAgentCallResponse

Translate high-level standardized agent-to-agent (A2A) message structures into backend-specific agent execution calls.

OpenAIChatCompletionRequest -> OpenAIChatCompletionResponse

Translate a standard OpenAI chat completion payload into a target LLM provider's native format, execute the network request, and map the response back to the OpenAI schema.

UserQuery, CitationGraph -> GroundedResponse

Ground LLM-generated responses by enriching queries with both semantic search matches and linked citation-graph nodes.

DocumentCorpus, FieldsetSchema -> StructuredExtractionGrid

Map a schema of natural-language queries into a parallelized extraction pipeline over a corpus of documents to produce structured tabular data.

Tuple<ToolRegistry, ToolCallRequest> -> ToolExecutionResult

Match an LLM's tool-call request against a registry of callable functions, execute the matched function, and return the result to the model context.

UnifiedPromptConfig -> UnifiedResponse

Translate a standardized generation config into provider-specific API requests and parse the responses back into a unified format.

AgentState -> NextAction

Generate the next logical action by prompting an LLM with current goals, history, and available tools.

UnifiedPrompt -> Stream<CompletionChunk>

Route inference tasks across heterogeneous local (e.g., Llama, Qwen) and cloud (e.g., OpenAI, Claude) engines using a normalized API interface.

BridgeConnection -> List<LLMToolDefinition>

Query a robotics middleware bridge to dynamically discover active topics, services, actions, and their schema types, translating them into structured LLM-callable tool definitions.

Requirement -> SoftwareArtifacts

Execute multi-step tasks by transitioning execution sequentially across specialized agent roles, passing upstream generated artifacts as downstream inputs.

Image + Prompt -> GroundedMultimodalResponse

Augment vision-language inference by dynamically triggering external search queries based on combined visual and textual input features.

ObjectDescription -> Mesh3D

Convert a text prompt or reference image of an object into a 3D mesh format suitable for physical simulation.

TextPrompt -> SpatialLayout

Convert a textual room description and embodied task into a symbolic spatial layout containing 3D object coordinate boundaries.

Query -> List<Entity>

Parse a natural language query using a language model to extract primary entities for graph seeding.