Package org.mockserver.llm.analysis

Class AgentRunAnalyzer

java.lang.Object

org.mockserver.llm.analysis.AgentRunAnalyzer

public class AgentRunAnalyzer
extends Object

Deterministic, read-only analysis of an agent run reconstructed from recorded LLM requests. Each request is decoded with the provider's ProviderCodec into a ParsedConversation; the richest conversation (most messages — i.e. the latest snapshot of the dialogue) is treated as the canonical run.

Pure (no network, no LLM): it inspects the structure the codecs already produce. Powers the verify_tool_call and explain_agent_run MCP tools.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final class	AgentRunAnalyzer.CallGraph	The correlated call graph of an agent run.
static final class	AgentRunAnalyzer.GraphEdge	A directed edge in the agent-run call graph.
static final class	AgentRunAnalyzer.GraphNode	A node in the agent-run call graph.
static final class	AgentRunAnalyzer.RunSummary	Structural summary of an agent run (for explain_agent_run).
static final class	AgentRunAnalyzer.ToolCallMatch	A single tool call the assistant made.
static final class	AgentRunAnalyzer.ToolCallReport	Result of a verify_tool_call-style inspection.

Constructor Summary

Constructors

Constructor	Description
AgentRunAnalyzer()

Method Summary

Modifier and Type	Method	Description
AgentRunAnalyzer.CallGraph	buildCallGraph(List<HttpRequest> requests, Provider provider)	Build a correlated call graph for an agent run: a node per message, a node per assistant tool call, NEXT edges along the message sequence, INVOKES edges from an assistant turn to the tool calls it made, and RESULT edges from a tool call to the tool message that returned its result (correlated by tool-call id, mirroring the matcher's correlation).
AgentRunAnalyzer.ToolCallReport	inspectToolCalls(List<HttpRequest> requests, Provider provider, String toolName, String argumentsRegex)	Count the assistant tool calls matching toolName (and, optionally, whose arguments match argumentsRegex) across the canonical conversation decoded from the given requests.
Optional<AgentRunAnalyzer.RunSummary>	summarise(List<HttpRequest> requests, Provider provider)	Summarise the canonical conversation's structure: message and assistant turn counts, the ordered sequence of tool-call names, the tool names a result was returned for, and the latest message's role.

Methods inherited from class java.lang.Object

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

Constructor Details

AgentRunAnalyzer

public AgentRunAnalyzer()

Method Details

inspectToolCalls

public AgentRunAnalyzer.ToolCallReport inspectToolCalls(List<HttpRequest> requests,
 Provider provider,
 String toolName,
 String argumentsRegex)

Count the assistant tool calls matching toolName (and, optionally, whose arguments match argumentsRegex) across the canonical conversation decoded from the given requests.

Parameters:

argumentsRegex - optional regex matched (via find) against the tool call's argument JSON string; null/empty means any

summarise

public Optional<AgentRunAnalyzer.RunSummary> summarise(List<HttpRequest> requests,
 Provider provider)

Summarise the canonical conversation's structure: message and assistant turn counts, the ordered sequence of tool-call names, the tool names a result was returned for, and the latest message's role. Returns empty when no conversation can be decoded.

buildCallGraph

public AgentRunAnalyzer.CallGraph buildCallGraph(List<HttpRequest> requests,
 Provider provider)

Build a correlated call graph for an agent run: a node per message, a node per assistant tool call, NEXT edges along the message sequence, INVOKES edges from an assistant turn to the tool calls it made, and RESULT edges from a tool call to the tool message that returned its result (correlated by tool-call id, mirroring the matcher's correlation). Empty graph when nothing decodes.