mini-swe-agent · SWE-agent team (Princeton/Stanford)
A deliberately radical ~100-line Python agent with bash as its only tool (no tool-calling API, no config sprawl) that still scores >74% on SWE-bench Verified, used by Meta, NVIDIA, and IBM as a baseline/benchmarking harness and proof that scaffolding can be minimal.
Identity
| Homepage | https://mini-swe-agent.com |
|---|---|
| Repository | https://github.com/SWE-agent/mini-swe-agent |
| License | open-source |
| Domain | coding |
| Form factor | cli |
Model support
| Model-agnostic | ✓ yes |
|---|---|
| Native models | — |
| Local models | ✓ yes |
| Pricing | byok |
Extensibility
| MCP client | ○ no |
|---|---|
| Agent Skills support | ○ no |
| Plugins / hooks | ○ no |
Execution & safety
| Sandboxing | optional |
|---|---|
| Autonomy spectrum | supervised → autonomous-local |
| Subagents | ○ no |
| Parallel sessions | ✓ yes |
| Context management | none-observed |
Integration & evidence
| Headless / API | ✓ yes |
|---|---|
| GitHub integration | ✓ yes |
| SWE-bench Verified | 74% |
| Sources |
✓ profile last verified 2026-07-08 · benchmark figures are vendor-reported unless a source states otherwise