View-oriented Conversation Compiler for Agent Trace Analysis

2026-03-31

AI Summary

This work addresses the limitations of unstructured agent interaction logs (e.g., in JSONL format), which significantly hinder in-context learning and trajectory analysis. To overcome this, the authors introduce the View-oriented Conversation Compiler (VCC)—the first application of compiler technology to agent trajectory analysis—which transforms raw logs into three structured views: complete, user-interface, and adaptive. VCC leverages a classical compilation pipeline comprising lexical analysis, syntactic parsing, intermediate representation, and code generation. Experiments on AppWorld demonstrate that simply replacing the raw input to the reflexion module with VCC-generated views consistently improves task success rates across three model configurations, while reducing reflexion token consumption by 50%–67% and yielding more concise and effective memory representations. This study establishes message formatting as a critical infrastructure for in-context learning, rather than a mere engineering detail.

Abstract

Agent traces carry increasing analytical value in the era of context learning and harness-driven agentic cognition, yet most prior work treats conversation format as a trivial engineering detail. Modern agent conversations contain deeply structured content, including nested tool calls and results, chain-of-thought reasoning blocks, sub-agent invocations, context-window compaction boundaries, and harness-injected system directives, whose complexity far exceeds that of simple user-assistant exchanges. Feeding such traces to a reflector or other analytical mechanism in plain text, JSON, YAML, or via grep can materially degrade analysis quality. This paper presents VCC (View-oriented Conversation Compiler), a compiler (lex, parse, IR, lower, emit) that transforms raw agent JSONL logs into a family of structured views: a full view (lossless transcript serving as the canonical line-number coordinate system), a user-interface view (reconstructing the interaction as the user actually perceived it), and an adaptive view (a structure-preserving projection governed by a relevance predicate). In a context-learning experiment on AppWorld, replacing only the reflector's input format, from raw JSONL to VCC-compiled views, leads to higher pass rates across all three model configurations tested, while cutting reflector token consumption by half to two-thirds and producing more concise learned memory. These results suggest that message format functions as infrastructure for context learning, not as an incidental implementation choice.

Problem

• agent trace analysis
• conversation format
• structured content
• context learning
• message representation

Innovation

• View-oriented Compilation
• Agent Trace Analysis
• Structured Conversation Views
• Context Learning Infrastructure
• Conversation Format Optimization