BootstrapAgent: Distilling Repository Setup into Reusable Agent Knowledge

This repository contains the official code for the paper:

BootstrapAgent: Distilling Repository Setup into Reusable Agent Knowledge

📄 Paper link: https://arxiv.org/abs/2605.15815

---

📌 Overview

Code agents increasingly help developers work with unfamiliar repositories, but every such task depends on a costly prerequisite: bootstrapping the repository into a usable development state. This process requires substantial trial-and-error exploration, yet the resulting knowledge--resolved dependencies, repair strategies--stays trapped in a single conversation, unavailable to future agents. We therefore formulate repository bootstrapping as a reusable startup knowledge problem and introduce BootstrapAgent, a multi-agent framework that distills the heuristics discovered during bootstrap exploration into a persistent, verifiable, agent-consumable .bootstrap contract. Through evidence extraction, structured planning, deterministic Docker-based verification, and trace-driven repair, BootstrapAgent generates a contract covering environment setup, diagnostic checks, minimal verification, and accumulated repair knowledge. We further propose warm repair with clean replay to accelerate iterative debugging without sacrificing cold-start reproducibility, and a delta repair with sanity check to prevent reward hacking. Experiments on three benchmarks show that BootstrapAgent achieves a 92.9% success rate, outperforming the baseline by over 10% while reducing downstream agent token usage by 25.9% and build time by 22.3%.

---

📂 Repository Layout

• code/: Python package, CLI, Docker verifier, benchmark runner, and tests.
• data/benchmark_inputs/: CSV inputs used to select benchmark repositories.
• REPRODUCIBILITY.md: environment setup, smoke tests, Docker verifier setup, and benchmark rerun notes.
• CITATION.cff: citation metadata placeholder to update once the arXiv identifier is assigned.
• LICENSE: MIT license for the released code and documentation.

Quick Start

cd code
python -m pip install -e '.[dev]'
python -m pytest

Optional LLM-agent dependencies can be installed with:

python -m pip install -e '.[agents,dev]'

Generate a .bootstrap/ package for a local repository without Docker verification:

rethink bootstrap --repo /path/to/repo --out runs/manual --allow-fallback --no-verify

Build the Docker verifier image from code/:

scripts/build-bootstrap-base-image.sh

Then verify a repository that already contains .bootstrap/:

rethink verify --repo /path/to/repo-with-bootstrap

Benchmark Inputs

The included CSV files are compact benchmark input lists, not full benchmark outputs. To prepare local benchmark checkouts:

cd code
python scripts/clone_benchmark_repos.py ../data/benchmark_inputs/repo2run_selected_122.csv --out /tmp/rethink-benchmark

To run a small row slice:

python scripts/run_csv_slice.py ../data/benchmark_inputs/repo2run_selected_122.csv 1 3 \
  --benchmark /tmp/rethink-benchmark \
  --out runs/repo2run-slice \
  --allow-fallback \
  --no-verify

Full benchmark reproduction may require Docker, network access, local storage for cloned repositories, and provider API keys for non-fallback LLM runs. See REPRODUCIBILITY.md for the longer protocol.

📄 Citation

@misc{fu2026bootstrapagentdistillingrepositorysetup,
      title={BootstrapAgent: Distilling Repository Setup into Reusable Agent Knowledge}, 
      author={Sihan Fu and Oucheng Liu and Shiyuan Wang and Jin Shi and Chengkun Wei},
      year={2026},
      eprint={2605.15815},
      archivePrefix={arXiv},
      primaryClass={cs.SE},
      url={https://arxiv.org/abs/2605.15815}, 
}