How Should Agents See Code? Rethinking Code Representations for Agentic Software Engineering
Melika Sepidband, Jiho Shin, Song Wang and Hadi Hemmati (York University) expose five repository maps (AST, call graph, module dependencies, documentation-enriched search and a hybrid) as MCP tools to Claude Code with Haiku 4.5 and compare them with raw file and shell access on four software engineering tasks.
Ask this paper
Fault localization. On LinuxFLBench all five maps improve recall@10 (0.757 to 0.829 vs 0.717) and cut token use by 25% to 39%.
Program repair. On SWE-bench Pro, resolution stays near the baseline (44.3% to 50.6% vs 49.0%) while tokens drop 17% to 26%.
Security patching. On SecBench the call graph gives the best patch rate in this sample (34.3% vs 28.9%) with fewer tokens, but every map increases latency.
Test generation. On SWT-bench, Hybrid and Docs reach 47.8% and 47.2% reproducing tests against 44.7%, while all maps use more tokens and time.
Guidance. Structured maps help most when finding the right code is the bottleneck. Raw search remains better for exact strings and non-code files, so the representation should be chosen per task.
Abstract
LLM agents are increasingly used for repository-level software engineering tasks. However, most agents still access repositories through raw files, lexical search, and shell commands, treating code primarily as text. This paper investigates map-based code-representation workflows as a design variable in agentic software engineering, examining how different repository views change agents' retrieval behavior, task success, token use, and latency. In this paper, we study five code-representation workflows: abstract syntax trees, call graphs, module dependencies, documentation-enriched search, and a hybrid workflow that combines them. Each workflow exposes a repository map that the agent can query during a task. We evaluate Claude Code with Haiku 4.5 against a raw-tool baseline on four repository-level tasks: fault localization with LinuxFLBench, repository-level code generation and repair with SWE-bench Pro, test generation with SWT-bench, and security patching with SecBench. The results show that representation workflows are most useful as token-saving tools when localization is the bottleneck, while their success gains narrow when the decisive step is patch or test authoring. On LinuxFLBench, all five workflows improve recall@10 over the baseline (0.757-0.829 vs. 0.717) and reduce token usage by 25-39%. On SWE-bench Pro, resolution remains close to the baseline (44.3-50.6% vs. 49.0%) while token usage drops by 17-26%. On SecBench, Callgraph has the highest patch-resolution rate in this sample (34.3% vs. 28.9%) and uses fewer tokens, but all workflows increase latency. On SWT-bench, Hybrid and Docs have the highest reproducing-test rates in this sample (47.8% and 47.2% vs. 44.7%), but all workflows use more tokens and take longer. Overall, map-based workflows change both what the agent inspects and how much it spends, but their benefits are task-dependent rather than universal.