All evidence cases
DeepSWE v1.1repositoryTypeScript

Implement recursive agent delegation through delegate_task tool calls

Run delegated sub-agents, feed their results back as tool_result, and handle unknown agents, failures, and circular delegation.

claude-code-by-agents-recursive-delegation

Normalized profile

What this task asks for

Run delegated sub-agents, feed their results back as tool_result, and handle unknown agents, failures, and circular delegation. Trigger delegation through `delegate_task` with an `agent_id` and delegated instructions. Run the selected sub-agent and feed one JSON `tool_result` containing its accumulated text back to the delegating agent. Match the streamed tool-use ID to `tool_result.tool_use_id`. Report unknown agents with both a stream error and an error tool result that names the requested agent. Report sub-agent failures only through an error tool result. Report circular delegation with a stream-level error whose message mentions `circular`.

  • feature implementation
  • network service
  • Claude Code by Agents
  • JSON
  • repository patch
  • recursive delegation and cross-agent error propagation

Published evidence

Model × harness outcomes

Counts are tied to these exact configurations. Harnesses are not treated as equivalent across benchmarks.

Models
13
Configurations
41
Trials
163

Showing 41 of 41 configurations

ModelHarnessEffortResultPasses
claude-fable-5vertex_aimini-swe-agentpublished configurationlowMixed2/4
claude-fable-5vertex_aimini-swe-agentpublished configurationmediumMixed1/4
claude-fable-5vertex_aimini-swe-agentpublished configurationhighMixed1/4
claude-fable-5vertex_aimini-swe-agentpublished configurationxhighMixed1/4
claude-fable-5vertex_aimini-swe-agentpublished configurationmaxNo passes0/4
claude-opus-4-8anthropicmini-swe-agentpublished configurationlowMixed3/4
claude-opus-4-8anthropicmini-swe-agentpublished configurationmediumMixed3/4
claude-opus-4-8anthropicmini-swe-agentpublished configurationhighMixed1/4
claude-opus-4-8anthropicmini-swe-agentpublished configurationxhighMixed1/4
claude-opus-4-8anthropicmini-swe-agentpublished configurationmaxMixed1/4
claude-sonnet-4-6anthropicmini-swe-agentpublished configurationhighMixed3/4
claude-sonnet-5anthropicmini-swe-agentpublished configurationlowMixed3/4
claude-sonnet-5anthropicmini-swe-agentpublished configurationmediumNo passes0/4
claude-sonnet-5anthropicmini-swe-agentpublished configurationhighMixed1/4
claude-sonnet-5anthropicmini-swe-agentpublished configurationxhighMixed1/4
claude-sonnet-5anthropicmini-swe-agentpublished configurationmaxNo passes0/4
gemini-3-1-pro-previewvertex_aimini-swe-agentpublished configurationhighMixed2/4
gemini-3-5-flashvertex_aimini-swe-agentpublished configurationmediumMixed3/4
glm-5-2zaimini-swe-agentpublished configurationhighMixed2/4
glm-5-2zaimini-swe-agentpublished configurationmaxMixed2/4
gpt-5-4openaimini-swe-agentpublished configurationxhighAll passed4/4
gpt-5-5openaimini-swe-agentpublished configurationlowMixed1/4
gpt-5-5openaimini-swe-agentpublished configurationmediumMixed1/4
gpt-5-5openaimini-swe-agentpublished configurationhighMixed1/4
gpt-5-5openaimini-swe-agentpublished configurationxhighMixed2/4
gpt-5-6-lunaopenaimini-swe-agentpublished configurationlowMixed2/4
gpt-5-6-lunaopenaimini-swe-agentpublished configurationmediumMixed1/4
gpt-5-6-lunaopenaimini-swe-agentpublished configurationhighMixed2/4
gpt-5-6-lunaopenaimini-swe-agentpublished configurationxhighMixed2/4
gpt-5-6-lunaopenaimini-swe-agentpublished configurationmaxAll passed4/4
gpt-5-6-solopenaimini-swe-agentpublished configurationlowAll passed4/4
gpt-5-6-solopenaimini-swe-agentpublished configurationmediumAll passed4/4
gpt-5-6-solopenaimini-swe-agentpublished configurationhighAll passed4/4
gpt-5-6-solopenaimini-swe-agentpublished configurationxhighMixed3/4
gpt-5-6-solopenaimini-swe-agentpublished configurationmaxMixed2/4
gpt-5-6-terraopenaimini-swe-agentpublished configurationlowAll passed3/3
gpt-5-6-terraopenaimini-swe-agentpublished configurationmediumAll passed4/4
gpt-5-6-terraopenaimini-swe-agentpublished configurationhighAll passed4/4
gpt-5-6-terraopenaimini-swe-agentpublished configurationxhighAll passed4/4
gpt-5-6-terraopenaimini-swe-agentpublished configurationmaxAll passed4/4
kimi-k2-7-codeopenroutermini-swe-agentpublished configurationdefaultMixed2/4