ProjAgent: Procedural Similarity Retrieval for Repository-Level Code Generation

2026-07-09 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, quick

Summary

ProjAgent is a novel repository-level code generation system designed to address complex cross-file dependencies and project-specific conventions. It introduces procedural similarity as an explicit retrieval signal, moving beyond traditional lexical, structural, or semantic methods. ProjAgent operates by decomposing target functions into intermediate reasoning steps, then employing an agentic workflow to retrieve repository functions exhibiting similar procedural behavior at each stage. This procedural context is integrated with conventional semantic retrieval to build a richer repository context for code generation. Furthermore, ProjAgent incorporates a conservative static-analysis feedback loop, iteratively repairing generated code using compiler and static-analysis feedback. Evaluated on REPOCOD, ProjAgent achieved 41.14% Pass@1, outperforming existing retrieval-based baselines and demonstrating the effectiveness of procedural similarity as a new retrieval dimension.

Key takeaway

For AI Engineers developing repository-level code generation systems, integrating procedural similarity retrieval offers a significant performance uplift. You should consider adopting an agentic workflow to decompose target functions and retrieve context based on procedural behavior, complementing existing semantic methods. Additionally, incorporating a static-analysis feedback loop can iteratively refine generated code, improving overall reliability and reducing manual debugging efforts.

Key insights

Procedural similarity, not just lexical or semantic, enhances repository-level code generation by matching functional logic.

Principles

• Decompose target functions into intermediate reasoning steps.
• Integrate procedural and semantic retrieval for richer context.
• Employ static-analysis feedback for iterative code repair.

Method

ProjAgent decomposes target functions, uses an agentic workflow to retrieve procedurally similar repository functions at each step, integrates this with semantic retrieval, and iteratively repairs code via static-analysis feedback.

In practice

• Apply agentic workflows for step-by-step code generation.
• Combine multiple retrieval signals for context enrichment.
• Implement static analysis for automated code refinement.

Topics

• Repository-level Code Generation
• Procedural Similarity
• Agentic AI
• Code Retrieval
• Static Analysis
• REPOCOD Benchmark

Best for: Research Scientist, AI Scientist, AI Engineer, Machine Learning Engineer

Related on AIssential

• ProjAgent: Procedural Similarity Retrieval for Repository-Level Code Generation — Procedural similarity, identified via LLM hidden-state projections, significantly enhances repository-level code generation.
• Remember Your Trace: Memory-Guided Long-Horizon Agentic Framework for Consistent and Hierarchical Repository-Level Code Documentation — Long-horizon agentic frameworks with shared memory improve code documentation consistency and efficiency.
• DeNovoSWE: Scaling Long-Horizon Environments for Generating Entire Repositories from Scratch — Automated, structured data generation with difficulty-aware filtering scales long-horizon software engineering training for LLM agents.
• On Accelerating Grounded Code Development for Research — Lexical search and structured workflows can effectively ground coding agents in dynamic, domain-specific research.
• Agent trajectories as programs: fingerprinting and programming coding-agent behavior — Agent problem-solving trajectories can be fingerprinted and analyzed procedurally to reveal distinct behavioral patterns beyond mere success rates.
• Multi-Perspective Agentic Program Repair via Code Property Graphs and Temporal Execution Graphs — Multi-perspective agentic reasoning with structured code and execution graphs significantly improves automated program repair.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.