Patterns

InboundEnvelope -> OutboundEnvelope

Orchestrate an agent's turn sequence through a cascade of interceptable lifecycle hooks (resolve, load, build prompt, run model, save, render, dispatch).

ToolOutputImage -> TextDescription

Convert visual outputs produced by execution tools into descriptive text summaries using a Vision-Language Model.

TaskSpecification -> StandardizedExecutionStream

Takes disparate LLM developer client tools -> produces a standardized execution protocol for unified task assignment and monitoring.

TaskPrompt -> List<SubAgentResult>

Decompose a complex parent prompt into parallel child-agent subtasks and aggregate their outputs back into the primary thread.

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.

PendingAction -> WorkflowResumeSignal

Pause agent execution on sensitive actions to request frontend confirmation, resuming only upon receipt of user approval.

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.

ToolCallInstruction -> ToolOutputData

Executes external actions through Model Context Protocol (MCP) compatible plugins based on model-generated tool calls.

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.

ActionPayload -> ActionResult

Expose dynamic, unregistered application commands via static proxy execution tools to run arbitrary actions without registry bloat.

AgentState -> NextAction

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