SWE-agent
Princeton's autonomous agent that resolves real GitHub issues end-to-end using a custom agent-computer interface.
SWE-agent
SWE-agent from Princeton NLP turns any LLM into a software engineering agent capable of finding and fixing bugs in real-world GitHub repositories. It uses a carefully designed Agent-Computer Interface (ACI) that gives the model specialised commands for navigating codebases — dramatically improving performance over raw tool-use.
Key Features
- Agent-Computer Interface: custom shell commands (
search_file,view_file,edit_file) optimised for LLM code navigation - SWE-bench compatible: benchmarked on the full SWE-bench Verified dataset; reproducible evaluation scripts included
- Multi-model: GPT-4o, Claude, Gemini, and any LiteLLM endpoint
- Batch mode: run the agent over a dataset of issues in parallel for evaluation
- Web UI: interactive traj viewer to replay and debug agent trajectories step-by-step
- Extensible: plug in custom tools, filters, and cost controls
Quick Start
pip install sweagent
# Fix a GitHub issue
swe-agent run \
--model_name claude-sonnet-4-5 \
--data_path https://github.com/owner/repo/issues/42 \
--repo_path /path/to/local/repo
npx ai-supply add swe-agent-github-issue-solver
Curated mirror of the open-source SWE-agent (MIT). Get it from the source.
Compromise signals — malicious or tampered code (leaked secrets, backdoors, a dropped executable) — reduce the score, and known dependency CVEs carry a bounded penalty (they warrant review but never QUARANTINE — update the dependency to clear). Other dangerous-by-capability traits are risk surface, expected for some capabilities. Every finding is mapped to its OWASP control below.
Findings mapped to the OWASP Top 10 for LLM Applications (2025) and the OWASP Machine Learning Security Top 10. Expand any flagged control for the exact findings — compromise reduces the score; expected/risk-surface do not, except a known CVE, which carries a small bounded penalty (high/critical → Review).
The same gate an agent runs before installing (POST /api/v1/trust/swe-agent-github-issue-solver/check). Click a policy:
Consume SWE-agent programmatically. Authenticate with an API key or session — see Authorize an agent.
# Agents: CHECK BEFORE YOU INSTALL (no auth) — score, grade, level, capability manifest
curl https://ai-supply.store/api/v1/trust/swe-agent-github-issue-solver
# Gate against your org policy (returns { pass, violations })
curl -X POST https://ai-supply.store/api/v1/trust/swe-agent-github-issue-solver/check \
-H "Content-Type: application/json" \
-d '{"minGrade":"B","denyPermissions":["shell"],"denyUnknownEgress":true}'
# CLI
npx ai-supply add swe-agent-github-issue-solver
# REST (install → download)
curl -X POST https://ai-supply.store/api/v1/listings/swe-agent-github-issue-solver/install \
-H "Authorization: Bearer $AIM_KEY"
# MCP tool
install_listing({ "slug": "swe-agent-github-issue-solver" })OpenAPI spec →Curated mirror — latest upstream source. See the repository for tagged releases.