We apply for the following ACM artifact badges (Artifact Review and Badging Version 1.1):
- Artifacts Available
- Artifacts Evaluated — Functional
- Artifacts Evaluated — Reusable
Author-created artifacts for this paper are publicly archived on Zenodo with a persistent DOI:
The deposit includes results.zip (translations, ablation agents, trajectories, graphectory analysis, cost data) and results.xlsx (detailed per-tool and per-project metrics). This repository (source code, Docker packaging, scripts, and documentation) is intended to be obtained via the same archival deposit so the complete artifact is available under a DOI.
The artifact is documented, consistent with the paper, complete, and exercisable, as described in README.md:
- Getting Started — artifact description, Docker installation, Zenodo results, credentials, and project structure
- Step-by-step instructions — reproduce RQ1–RQ4 via
scripts/start_docker.sh,scripts/run_test_comparison.sh,src/analysis/ablation.py, andsrc/analysis/cost.py, using either fresh runs or unpacked Zenodoresults.zipunderresults/ - Docker packaging under
docker-env/installs dependencies for the supported language pairs
LLM API keys are not included; users configure Bedrock or OpenRouter credentials as in README.md § Credentials. Empirical outputs matching the paper are also provided on Zenodo.
The artifact is carefully structured for reuse and extension beyond minimal functionality:
- Modular multi-agent pipeline (
src/agents/recodeagent/) and baselines (src/agents/baseagent/) - MCP servers and analysis utilities (
src/mcp/,src/analysis/) - Docker environment covering six PLs and four PL pairs (crust, oxidizer, alphatrans, skel)
- Documented guidance for new projects, languages, and MCP servers (
README.md§ Building on ReCodeAgent)