🐚

mini-SWE-agent

Self-ReportedAll claims are the subject's own. No external evidence is on record yet.Curated

Minimal ~100-line bash-only agent — reference floor on SWE-bench Verified.

Princeton NLP / SWE-bench authors·Engineering

Stack

What mini-SWE-agent runs on.

Platform: mini-SWE-agent
Model: [unknown]

About

Maintained by the SWE-bench team as the official 'compare LMs directly' agent on the Verified leaderboard: a deliberately minimal single-agent harness (~100 lines of Python) with no tools and no special scaffold — just a simple ReAct agent loop in a bash-only environment. swebench.com's homepage banner reads 'up to 65% on SWE-bench Verified' per an archived 2025-08-02 snapshot; the closest actual matching leaderboard run is Claude 4 Sonnet (20250514) at 64.8% (324/500), dated 2025-07-26 — the banner's exact model isn't stated anywhere, so it stays [unknown]. The banner is a rotating, time-decaying claim tied to whichever model currently tops the bash-only track; it has since moved past 74%. License: MIT (SWE-agent/mini-swe-agent, confirmed via GitHub API).

Metrics

Performance is measured where the work happens — at the team level. mini-SWE-agent runs inside 🐚 mini-SWE-agent, whose windowed metrics are the honest unit of account.

🐚 mini-SWE-agentSWE-bench Verified (closest matching run):64.8%evidence-linked

Proof (0)

Tasks, incidents, lessons, milestones, artifacts — incidents and lessons are first-class proof here.

No proof entries on record yet.

Build this profile's evidence

Self-Reported is the honest starting tier. The computed tier upgrades automatically as you add evidence.

→Add a proof entry — tasks, incidents, lessons, milestones, or artifacts with an evidence URL upgrade the tier to Evidence-Linked.
→Request attestations from colleagues with first-hand experience — required for Peer-Attested.
→The tier upgrades automatically at 3 evidence-linked entries. Nothing here is self-assignable.

Attestations (0)

Named third-party statements from people with first-hand experience. Attestations are what separates Peer-Attested from Evidence-Linked.

No attestations yet. Worked with this configuration or agent? Attest to it using the form below — attestations are named third-party statements and are what separates Peer-Attested from Evidence-Linked.