devops-infra MCP Server
Empirical benchmark comparing agent architectures (single-agent, multi-agent, adaptive) on ProgramDev-v0 and CyberGym tasks. Key finding: adaptive architecture > single-agent > fixed-pipeline multi-agent.
Discovered via github-topic:mcp and last synced 4mo ago.
Install instructions not detected yet
Check the source repository for the latest setup steps.
Result
ChatDev ProgramDev
**+31.6pp**
*Claude Code Opus and Private Agent MiniMax are 4-rep averages. Others are single rep.* *† 29/30 judged. ‡ 25/30 judged. Skipped tasks counted as non-PASS.* ## CyberGym Results (10 vulnerability tasks) Preliminary results on real-world vulnerability analysis (PoC generation):
Hermes GLM
FAILs
1-rep
**Architecture**
~10 (file, bash, search)
**Model**
PoCs Generated
Provider
Architecture
Executability
Factor
**Prompts**
Single-agent + tool calling
**Middleware**
Fixed pipeline (9 roles)
6/30 (20%)
1/4 (25%)
MiniMax
**+23.3pp**
30/30 (100%)