Coding Agent

Baseline wrapping Princeton's mini-swe-agent v2.2.8 with Claude Sonnet 4.6 via LiteLLM.

The Purple Coding Agent is a high-performance, autonomous software engineering agent optimized for repository-level reasoning and complex bug resolution in competitive environments like SWE-Bench Pro and AIMO2026. Operating on a stateful Phase 2 architecture, the agent moves beyond static code analysis by utilizing a live, execution-grounded environment. It autonomously explores codebases, reproduces issues within isolated Docker containers, and verifies its own repairs through a mechanical test gate to ensure production-grade reliability. Key Capabilities Stateful Bash REPL: Maintains a persistent, 50-turn interactive session that allows the agent to explore, edit, and verify code iteratively within a single unified context. Mechanical Ground Truth: Utilizes a Docker-out-of-Docker (DooD) bridge to spawn sibling containers, allowing it to run test suites natively and generate its own diagnostic logs. Inference-Time Scaling (GRPO): Employs group sampling strategies to generate and evaluate multiple diagnostic hypotheses simultaneously, prioritizing leads based on real-world execution feedback. Graph-Based RAG: Leverages Tree-Sitter for AST-based repository mapping, providing the agent with a structural "skeleton" of the codebase to prevent context wandering in large repositories. Relative Reward Verification: Implements a smarter QA gate that compares post-fix execution results against a baseline state to prevent regressions and ensure the core issue is resolved. Automated Tooling: Seamlessly integrates specialized models (e.g., DeepSeek-v4-flash) with local bash utilities to perform batched file reads and robust Python-based edits.

Codewalk Q&A Evaluator Agent benchmarks AI agents on their ability to help software engineers interact with a codebase, build understanding of its concepts, and contribute back. Given a question about a repository (e.g., "How does request processing work in FastAPI?"), the evaluator sends it to a Q&A agent via the A2A protocol, then uses an LLM judge to score the response on four dimensions: - Architecture-Level Reasoning (0-5) – Clear reasoning about system design, modules, and architecture - Reasoning Consistency (0-5) – Logical, coherent flow of explanation - Code Understanding Tier (0-5) – Depth of understanding from performance to architectural level - Grounding (0-5) – Factual accuracy and alignment with reference answers While currently evaluating against open-source repositories, the system supports closed-source codebases as well. The benchmark supports multiple judge models (Gemini, Claude etc) and is part of the broader Codewalk project, which aims to build AI that maintains deep understanding of codebases from multiple software engineering perspectives—architecture, reliability, maintainability, and beyond.

RLM-style purple agent for Terminal Bench 2.0. Root LM (Opus) drives a persistent in-process REPL with a context-offloaded transcript and a Haiku sub-LLM for filtering large outputs.