You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 42e4842
Browse filesBrowse the repository at this point in the historyBrowse files
Anthropic 在 **How we built our multi-agent research system**([Anthropic, 2025d](https://www.anthropic.com/engineering/built-multi-agent-research-system))中详细介绍了其 Research 功能的架构。
Anthropic 在 **How we built our multi-agent research system**([Anthropic, 2025d](https://www.anthropic.com/engineering/built-multi-agent-research-system))中详细介绍了其 Research 功能的架构。lead agent 将研究任务拆给并行运行的 search subagent,综合检索结果后,再由 citation agent 定位报告中的具体引用位置。
234
238
235
239
{{< figure
236
240
src="multi-agent-architecture.png"
@@ -239,21 +243,59 @@ Anthropic 在 **How we built our multi-agent research system**([Anthropic, 202
239
243
width="100%"
240
244
>}}
241
245
242
-
这是一个典型的 orchestrator-worker 模式的多智能体架构:由一个 lead agent 协调全程并把任务委派给并行运行的专业化 Subagents,最后由 CitationAgent 处理文档和 research report 以定位具体引用位置。完整的执行流程如下图所示:
caption="Fig. 15. The full process of the multi-agent research system: the LeadResearcher plans and spawns subagents, which retrieve in parallel and return results, and finally the CitationAgent aligns citation locations to produce the report. (Image source: [Anthropic, 2025d](https://www.anthropic.com/engineering/built-multi-agent-research-system))"
caption="Fig. 16. Coordinated swarm findings (solid curves) versus independently partitioned agents (stars); dashed curves show overlap, while the dotted Mythos Preview curve keeps only findings inside the baseline's core directories. Search scopes and token budgets are not matched. (Image source: [Anthropic, 2026e](https://www.anthropic.com/research/multiagent-systems))"
caption="Fig. 17. End-of-run merged-PR fraction and median-agent code sharing for 12-hour game-building swarms, averaged across three prompt types. These are coordination proxies rather than quality scores, and high merge rates can reflect file isolation instead of collaboration. (Image source: [Anthropic, 2026e](https://www.anthropic.com/research/multiagent-systems))"
caption="Fig. 16. Claude Code parallelizes work in several ways: subagents delegate a side task in their own context and return only a summary, agent teams coordinate through a shared task list with direct messaging, and dynamic workflows script and cross-check many subagents for jobs too big to coordinate one turn at a time. (Image source: [Cat Wu on X](https://x.com/_catwu/status/2060054180379689074))"
325
+
caption="Fig. 18. Claude Code parallelizes work in several ways: subagents delegate a side task in their own context and return only a summary, agent teams coordinate through a shared task list with direct messaging, and dynamic workflows script and cross-check many subagents for jobs too big to coordinate one turn at a time. (Image source: [Cat Wu on X](https://x.com/_catwu/status/2060054180379689074))"
caption="Fig. 17. Six common dynamic workflow patterns in Claude Code: classify-and-act, fan-out-and-synthesize, adversarial verification, generate-and-filter, tournament, and loop until done. (Image source: [Anthropic, 2026d](https://claude.com/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code))"
336
+
caption="Fig. 19. Six common dynamic workflow patterns in Claude Code: classify-and-act, fan-out-and-synthesize, adversarial verification, generate-and-filter, tournament, and loop until done. (Image source: [Anthropic, 2026d](https://claude.com/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code))"
caption="Fig. 18. The agent system architectures studied: a single-agent baseline alongside several multi-agent coordination topologies. (Image source: [Kim et al., 2025](https://arxiv.org/abs/2512.08296))"
358
+
caption="Fig. 20. The agent system architectures studied: a single-agent baseline alongside several multi-agent coordination topologies. (Image source: [Kim et al., 2025](https://arxiv.org/abs/2512.08296))"
caption="Fig. 19. Multi-agent coordination helps on parallelizable tasks but hurts on sequential ones, and the benefit shrinks as the single-agent baseline grows stronger (capability saturation). (Image source: [Kim et al., 2025](https://arxiv.org/abs/2512.08296))"
368
+
caption="Fig. 21. Multi-agent coordination helps on parallelizable tasks but hurts on sequential ones, and the benefit shrinks as the single-agent baseline grows stronger (capability saturation). (Image source: [Kim et al., 2025](https://arxiv.org/abs/2512.08296))"
[10] Anthropic. ["How We Built Our Multi-Agent Research System."](https://www.anthropic.com/engineering/built-multi-agent-research-system) Anthropic Engineering Blog (2025d).
389
433
390
-
[11] Anthropic. ["A Harness for Every Task: Dynamic Workflows in Claude Code."](https://claude.com/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code) Anthropic Blog (2026d).
434
+
[11] Phillips, Cara. ["Multi-Agent Coordination Patterns: Five Approaches and When to Use Them."](https://claude.com/blog/multi-agent-coordination-patterns) Claude Blog (Apr. 10, 2026).
435
+
436
+
[12] Anthropic Frontier Red Team. ["Patterns and Problems in Emerging Multiagent Systems."](https://www.anthropic.com/research/multiagent-systems) Anthropic Research (Aug. 13, 2026).
437
+
438
+
[13] Anthropic. ["A Harness for Every Task: Dynamic Workflows in Claude Code."](https://claude.com/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code) Anthropic Blog (2026d).
391
439
392
-
[12] Kim, Yubin, et al. ["Towards a Science of Scaling Agent Systems."](https://arxiv.org/abs/2512.08296) arXiv preprint arXiv:2512.08296 (2025).
440
+
[14] Kim, Yubin, et al. ["Towards a Science of Scaling Agent Systems."](https://arxiv.org/abs/2512.08296) arXiv preprint arXiv:2512.08296 (2025).
0 commit comments