Teams for bug fixing

Working harness designs — topology, agent roster, model choices, and the evidence behind them. 7 teams match.

1 operated · 6 reference designs — operated teams are listed first, reference designs after them.

See stack trends →

Run an agent team? Add it — no public links needed.

SortFeatured Recently active Highest trust

Next.js Maintainer Agent

Next.js

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 1 of the 3 evidence-linked proof entries needed for Evidence-Linked is on record.Curated

Independent agents the Next.js maintainers use to research, reproduce, verify and fix GitHub issues; they helped cut the backlog by 1,462 closed issues in about three weeks.

updated 2mo ago

Orchestrator–Worker7 listed membersCustom

Issue research agent· GPT-5.6 LunaReproduction agentVerification agentEnd-to-end test agent+3 more

software-delivery

Outcome[unknown]

Economics[unknown] · deliberate

MAGIS

co-evolve-lab

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 2 of the 3 evidence-linked proof entries needed for Evidence-Linked are on record.CuratedReference design

Agents coordinating repository changes for software evolution.

updated 19d ago

Orchestrator–Worker4 listed membersCustom

Process coordinatorRepository navigatorCode implementerCode reviewer

software-delivery

Outcome[unknown]

Economics[unknown] · deliberate

OpenHands (OpenDevin)

All Hands AI (OpenHands)

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 2 of the 3 evidence-linked proof entries needed for Evidence-Linked are on record.CuratedReference design

Open-source AI software developer with sandboxed runtime — ICLR 2025.

Solo + Tools1 agentOpenHands

software-delivery

Economics[unknown] · deliberate

Aider Architect/Editor

Aider (Paul Gauthier)

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 1 of the 3 evidence-linked proof entries needed for Evidence-Linked is on record.CuratedReference design

Two-role pipeline: reasoning architect + format-specialist editor — 85% on Aider's code-editing benchmark.

Pipeline2 agentsAider

Architect· o1-previewEditor· o1-mini

software-delivery

Economics[unknown] · deliberate

Claude SWE-Bench Team

Anthropic

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 1 of the 3 evidence-linked proof entries needed for Evidence-Linked is on record.CuratedReference design

Single-agent software engineer achieving 49% on SWE-bench Verified.

Solo + Tools1 agentClaude API

Software Engineer· Claude 3.5 Sonnet

software-delivery

Economics[unknown] · deliberate

Refact.ai Agent

Refact.ai

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 1 of the 3 evidence-linked proof entries needed for Evidence-Linked is on record.CuratedReference design

Solo agent, multi-model orchestration — 70.4% on SWE-bench Verified.

Solo + Tools1 agentRefact.ai

Solo Agent· Claude 3.7 Sonnet (multi-model orchestration)

software-delivery

Outcome[unknown]

Economics[unknown] · deliberate

SWE-agent (Princeton ACI)

Princeton NLP / SWE-bench authors

From public sources · Self-ReportedDocumented by AgentCV from cited public sources — not stated by the team. Tier: Self-Reported. 1 of the 3 evidence-linked proof entries needed for Evidence-Linked is on record.CuratedReference design

Solo software agent with custom ACI — 12.5% SWE-bench, 87.7% HumanEvalFix.

Solo + Tools1 agentCustom ACI (Docker)

Software Engineer

software-delivery

Economics[unknown] · deliberate